Metal: Implement Uniform buffers

README.md

@@ -10,8 +10,8 @@ Metal and MacOS, Chrome OS, and Fuchsia support.
 
 |                |  Direct3D 9   |  Direct3D 11     |   Desktop GL   |    GL ES      |    Vulkan     |    Metal      |
 |----------------|:-------------:|:----------------:|:--------------:|:-------------:|:-------------:|:-------------:|
-| OpenGL ES 2.0  |    complete   |    complete      |    complete    |    complete   |    complete   |  in progress  |
-| OpenGL ES 3.0  |               |    complete      |    complete    |    complete   |    complete   |               |
+| OpenGL ES 2.0  |    complete   |    complete      |    complete    |    complete   |    complete   |    complete   |
+| OpenGL ES 3.0  |               |    complete      |    complete    |    complete   |    complete   |  in progress  |
 | OpenGL ES 3.1  |               |   in progress    |    complete    |    complete   |    complete   |               |
 | OpenGL ES 3.2  |               |                  |  in progress   |  in progress  |  in progress  |               |
 

src/libANGLE/renderer/metal/BUILD.gn

@@ -35,6 +35,8 @@ _metal_backend_sources = [
   "SurfaceMtl.mm",
   "TextureMtl.h",
   "TextureMtl.mm",
+  "TransformFeedbackMtl.h",
+  "TransformFeedbackMtl.mm",
   "VertexArrayMtl.h",
   "VertexArrayMtl.mm",
   "mtl_buffer_pool.h",

src/libANGLE/renderer/metal/BufferMtl.h

@@ -63,6 +63,13 @@ struct IndexConversionBufferMtl : public ConversionBufferMtl
     const size_t offset;
 };
 
+struct UniformConversionBufferMtl : public ConversionBufferMtl
+{
+    UniformConversionBufferMtl(ContextMtl *context, size_t offsetIn);
+
+    const size_t offset;
+};
+
 class BufferHolderMtl
 {
   public:
@@ -138,6 +145,8 @@ class BufferMtl : public BufferImpl, public BufferHolderMtl
                                                        gl::DrawElementsType elemType,
                                                        size_t offset);
 
+    ConversionBufferMtl *getUniformConversionBuffer(ContextMtl *context, size_t offset);
+
     size_t size() const { return static_cast<size_t>(mState.getSize()); }
 
   private:
@@ -166,10 +175,12 @@ class BufferMtl : public BufferImpl, public BufferHolderMtl
     // GPU side buffers pool
     mtl::BufferPool mBufferPool;
 
-    // A cache of converted vertex data.
+    // A cache of converted buffer data.
     std::vector<VertexConversionBufferMtl> mVertexConversionBuffers;
 
     std::vector<IndexConversionBufferMtl> mIndexConversionBuffers;
+
+    std::vector<UniformConversionBufferMtl> mUniformConversionBuffers;
 };
 
 class SimpleWeakBufferHolderMtl : public BufferHolderMtl

src/libANGLE/renderer/metal/BufferMtl.mm

@@ -63,6 +63,12 @@ IndexConversionBufferMtl::IndexConversionBufferMtl(ContextMtl *context,
       offset(offsetIn)
 {}
 
+// UniformConversionBufferMtl implementation
+UniformConversionBufferMtl::UniformConversionBufferMtl(ContextMtl *context, size_t offsetIn)
+    : ConversionBufferMtl(context, 0, mtl::kUniformBufferSettingOffsetMinAlignment),
+      offset(offsetIn)
+{}
+
 // VertexConversionBufferMtl implementation.
 VertexConversionBufferMtl::VertexConversionBufferMtl(ContextMtl *context,
                                                      angle::FormatID formatIDIn,
@@ -339,6 +345,20 @@ IndexConversionBufferMtl *BufferMtl::getIndexConversionBuffer(ContextMtl *contex
     return &mIndexConversionBuffers.back();
 }
 
+ConversionBufferMtl *BufferMtl::getUniformConversionBuffer(ContextMtl *context, size_t offset)
+{
+    for (UniformConversionBufferMtl &buffer : mUniformConversionBuffers)
+    {
+        if (buffer.offset == offset)
+        {
+            return &buffer;
+        }
+    }
+
+    mUniformConversionBuffers.emplace_back(context, offset);
+    return &mUniformConversionBuffers.back();
+}
+
 void BufferMtl::markConversionBuffersDirty()
 {
     for (VertexConversionBufferMtl &buffer : mVertexConversionBuffers)
@@ -354,12 +374,20 @@ void BufferMtl::markConversionBuffersDirty()
         buffer.convertedBuffer = nullptr;
         buffer.convertedOffset = 0;
     }
+
+    for (UniformConversionBufferMtl &buffer : mUniformConversionBuffers)
+    {
+        buffer.dirty           = true;
+        buffer.convertedBuffer = nullptr;
+        buffer.convertedOffset = 0;
+    }
 }
 
 void BufferMtl::clearConversionBuffers()
 {
     mVertexConversionBuffers.clear();
     mIndexConversionBuffers.clear();
+    mUniformConversionBuffers.clear();
 }
 
 angle::Result BufferMtl::setDataImpl(const gl::Context *context,

src/libANGLE/renderer/metal/ContextMtl.h

@@ -420,7 +420,7 @@ class ContextMtl : public ContextImpl, public mtl::Context
     angle::Result handleDirtyDepthBias(const gl::Context *context);
     angle::Result checkIfPipelineChanged(const gl::Context *context,
                                          gl::PrimitiveMode primitiveMode,
-                                         Optional<mtl::RenderPipelineDesc> *changedPipelineDesc);
+                                         bool *isPipelineDescChanged);
 
     angle::Result startOcclusionQueryInRenderPass(QueryMtl *query, bool clearOldValue);
 
@@ -440,6 +440,7 @@ class ContextMtl : public ContextImpl, public mtl::Context
         DIRTY_BIT_CULL_MODE,
         DIRTY_BIT_WINDING,
         DIRTY_BIT_RENDER_PIPELINE,
+        DIRTY_BIT_UNIFORM_BUFFERS_BINDING,
         DIRTY_BIT_MAX,
     };
 

src/libANGLE/renderer/metal/ContextMtl.mm

@@ -22,6 +22,7 @@
 #include "libANGLE/renderer/metal/RenderTargetMtl.h"
 #include "libANGLE/renderer/metal/ShaderMtl.h"
 #include "libANGLE/renderer/metal/TextureMtl.h"
+#include "libANGLE/renderer/metal/TransformFeedbackMtl.h"
 #include "libANGLE/renderer/metal/VertexArrayMtl.h"
 #include "libANGLE/renderer/metal/mtl_command_buffer.h"
 #include "libANGLE/renderer/metal/mtl_format_utils.h"
@@ -795,7 +796,7 @@ angle::Result ContextMtl::syncState(const gl::Context *context,
                 // NOTE(hqle): ES 3.0 feature.
                 break;
             case gl::State::DIRTY_BIT_UNIFORM_BUFFER_BINDINGS:
-                // NOTE(hqle): ES 3.0 feature.
+                mDirtyBits.set(DIRTY_BIT_UNIFORM_BUFFERS_BINDING);
                 break;
             case gl::State::DIRTY_BIT_ATOMIC_COUNTER_BUFFER_BINDING:
                 break;
@@ -946,8 +947,7 @@ SyncImpl *ContextMtl::createSync()
 TransformFeedbackImpl *ContextMtl::createTransformFeedback(const gl::TransformFeedbackState &state)
 {
     // NOTE(hqle): ES 3.0
-    UNIMPLEMENTED();
-    return nullptr;
+    return new TransformFeedbackMtl(state);
 }
 
 // Sampler object creation
@@ -1644,8 +1644,10 @@ angle::Result ContextMtl::setupDraw(const gl::Context *context,
         ANGLE_TRY(startOcclusionQueryInRenderPass(mOcclusionQuery, false));
     }
 
-    Optional<mtl::RenderPipelineDesc> changedPipelineDesc;
-    ANGLE_TRY(checkIfPipelineChanged(context, mode, &changedPipelineDesc));
+    bool isPipelineDescChanged;
+    ANGLE_TRY(checkIfPipelineChanged(context, mode, &isPipelineDescChanged));
+
+    bool uniformBuffersDirty = false;
 
     for (size_t bit : mDirtyBits)
     {
@@ -1692,6 +1694,9 @@ angle::Result ContextMtl::setupDraw(const gl::Context *context,
             case DIRTY_BIT_RENDER_PIPELINE:
                 // Already handled. See checkIfPipelineChanged().
                 break;
+            case DIRTY_BIT_UNIFORM_BUFFERS_BINDING:
+                uniformBuffersDirty = true;
+                break;
             default:
                 UNREACHABLE();
                 break;
@@ -1700,7 +1705,8 @@ angle::Result ContextMtl::setupDraw(const gl::Context *context,
 
     mDirtyBits.reset();
 
-    ANGLE_TRY(mProgram->setupDraw(context, &mRenderEncoder, changedPipelineDesc, textureChanged));
+    ANGLE_TRY(mProgram->setupDraw(context, &mRenderEncoder, mRenderPipelineDesc,
+                                  isPipelineDescChanged, textureChanged, uniformBuffersDirty));
 
     if (mode == gl::PrimitiveMode::LineLoop)
     {
@@ -1892,10 +1898,9 @@ angle::Result ContextMtl::handleDirtyDepthBias(const gl::Context *context)
     return angle::Result::Continue;
 }
 
-angle::Result ContextMtl::checkIfPipelineChanged(
-    const gl::Context *context,
-    gl::PrimitiveMode primitiveMode,
-    Optional<mtl::RenderPipelineDesc> *changedPipelineDesc)
+angle::Result ContextMtl::checkIfPipelineChanged(const gl::Context *context,
+                                                 gl::PrimitiveMode primitiveMode,
+                                                 bool *isPipelineDescChanged)
 {
     ASSERT(mRenderEncoder.valid());
     mtl::PrimitiveTopologyClass topologyClass = mtl::GetPrimitiveTopologyClass(primitiveMode);
@@ -1920,10 +1925,10 @@ angle::Result ContextMtl::checkIfPipelineChanged(
 
         mRenderPipelineDesc.outputDescriptor.updateEnabledDrawBuffers(
             mDrawFramebuffer->getState().getEnabledDrawBuffers());
-
-        *changedPipelineDesc = mRenderPipelineDesc;
     }
 
+    *isPipelineDescChanged = rppChange;
+
     return angle::Result::Continue;
 }
 }

src/libANGLE/renderer/metal/DisplayMtl.mm

@@ -521,21 +521,20 @@ void DisplayMtl::ensureCapsInitialized() const
     mNativeCaps.fragmentMediumpInt.setTwosComplementInt(32);
     mNativeCaps.fragmentLowpInt.setTwosComplementInt(32);
 
-    GLuint maxUniformVectors = mtl::kDefaultUniformsMaxSize / (sizeof(GLfloat) * 4);
+    GLuint maxDefaultUniformVectors = mtl::kDefaultUniformsMaxSize / (sizeof(GLfloat) * 4);
 
-    const GLuint maxUniformComponents = maxUniformVectors * 4;
+    const GLuint maxDefaultUniformComponents = maxDefaultUniformVectors * 4;
 
     // Uniforms are implemented using a uniform buffer, so the max number of uniforms we can
     // support is the max buffer range divided by the size of a single uniform (4X float).
-    mNativeCaps.maxVertexUniformVectors                              = maxUniformVectors;
-    mNativeCaps.maxShaderUniformComponents[gl::ShaderType::Vertex]   = maxUniformComponents;
-    mNativeCaps.maxFragmentUniformVectors                            = maxUniformVectors;
-    mNativeCaps.maxShaderUniformComponents[gl::ShaderType::Fragment] = maxUniformComponents;
+    mNativeCaps.maxVertexUniformVectors                              = maxDefaultUniformVectors;
+    mNativeCaps.maxShaderUniformComponents[gl::ShaderType::Vertex]   = maxDefaultUniformComponents;
+    mNativeCaps.maxFragmentUniformVectors                            = maxDefaultUniformVectors;
+    mNativeCaps.maxShaderUniformComponents[gl::ShaderType::Fragment] = maxDefaultUniformComponents;
 
-    // NOTE(hqle): support UBO (ES 3.0 feature)
-    mNativeCaps.maxShaderUniformBlocks[gl::ShaderType::Vertex]   = 0;
-    mNativeCaps.maxShaderUniformBlocks[gl::ShaderType::Fragment] = 0;
-    mNativeCaps.maxCombinedUniformBlocks                         = 0;
+    mNativeCaps.maxShaderUniformBlocks[gl::ShaderType::Vertex]   = mtl::kMaxShaderUBOs;
+    mNativeCaps.maxShaderUniformBlocks[gl::ShaderType::Fragment] = mtl::kMaxShaderUBOs;
+    mNativeCaps.maxCombinedUniformBlocks                         = mtl::kMaxGLUBOBindings;
 
     // Note that we currently implement textures as combined image+samplers, so the limit is
     // the minimum of supported samplers and sampled images.
@@ -550,18 +549,20 @@ void DisplayMtl::ensureCapsInitialized() const
     mNativeCaps.maxCombinedShaderStorageBlocks                   = maxPerStageStorageBuffers;
 
     // Fill in additional limits for UBOs and SSBOs.
-    mNativeCaps.maxUniformBufferBindings     = 0;
-    mNativeCaps.maxUniformBlockSize          = 0;
-    mNativeCaps.uniformBufferOffsetAlignment = 0;
+    mNativeCaps.maxUniformBufferBindings = mNativeCaps.maxCombinedUniformBlocks;
+    mNativeCaps.maxUniformBlockSize      = mtl::kMaxUBOSize;  // Default according to GLES 3.0 spec.
+    mNativeCaps.uniformBufferOffsetAlignment = 1;
 
     mNativeCaps.maxShaderStorageBufferBindings     = 0;
     mNativeCaps.maxShaderStorageBlockSize          = 0;
     mNativeCaps.shaderStorageBufferOffsetAlignment = 0;
 
     // NOTE(hqle): support UBO
+    const uint32_t maxCombinedUniformComponents =
+        maxDefaultUniformComponents + mtl::kMaxUBOSize * mtl::kMaxShaderUBOs / 4;
     for (gl::ShaderType shaderType : gl::kAllGraphicsShaderTypes)
     {
-        mNativeCaps.maxCombinedShaderUniformComponents[shaderType] = maxUniformComponents;
+        mNativeCaps.maxCombinedShaderUniformComponents[shaderType] = maxCombinedUniformComponents;
     }
 
     mNativeCaps.maxCombinedShaderOutputResources = 0;

src/libANGLE/renderer/metal/ProgramMtl.h

@@ -18,6 +18,7 @@
 #include "common/utilities.h"
 #include "libANGLE/renderer/ProgramImpl.h"
 #include "libANGLE/renderer/glslang_wrapper_utils.h"
+#include "libANGLE/renderer/metal/mtl_buffer_pool.h"
 #include "libANGLE/renderer/metal/mtl_command_buffer.h"
 #include "libANGLE/renderer/metal/mtl_glslang_utils.h"
 #include "libANGLE/renderer/metal/mtl_resources.h"
@@ -27,15 +28,28 @@ namespace rx
 {
 class ContextMtl;
 
+struct ProgramArgumentBufferEncoderMtl
+{
+    void reset(ContextMtl *contextMtl);
+
+    mtl::AutoObjCPtr<id<MTLArgumentEncoder>> metalArgBufferEncoder;
+    mtl::BufferPool bufferPool;
+};
+
 // Represents a specialized shader variant. For example, a shader variant with fragment coverage
 // mask enabled and a shader variant without.
-struct ProgramShaderVariantMtl
+struct ProgramShaderObjVariantMtl
 {
     void reset(ContextMtl *contextMtl);
 
     mtl::AutoObjCPtr<id<MTLFunction>> metalShader;
-    // NOTE(hqle): might need additional info such as uniform buffer encoder, fragment coverage mask
-    // enabled or not, etc.
+    // UBO's argument buffer encoder. Used when number of UBOs used exceeds number of allowed
+    // discrete slots, and thus needs to encode all into one argument buffer.
+    ProgramArgumentBufferEncoderMtl uboArgBufferEncoder;
+
+    // Store reference to the TranslatedShaderInfo to easy querying mapped textures/UBO/XFB
+    // bindings.
+    const mtl::TranslatedShaderInfo *translatedSrcInfo;
 };
 
 class ProgramMtl : public ProgramImpl, public mtl::RenderPipelineCacheSpecializeShaderFactory
@@ -124,8 +138,10 @@ class ProgramMtl : public ProgramImpl, public mtl::RenderPipelineCacheSpecialize
     // shader program changed.
     angle::Result setupDraw(const gl::Context *glContext,
                             mtl::RenderCommandEncoder *cmdEncoder,
-                            const Optional<mtl::RenderPipelineDesc> &changedPipelineDesc,
-                            bool forceTexturesSetting);
+                            const mtl::RenderPipelineDesc &pipelineDesc,
+                            bool pipelineDescChanged,
+                            bool forceTexturesSetting,
+                            bool uniformBuffersDirty);
 
   private:
     template <int cols, int rows>
@@ -146,16 +162,32 @@ class ProgramMtl : public ProgramImpl, public mtl::RenderPipelineCacheSpecialize
                                  mtl::RenderCommandEncoder *cmdEncoder,
                                  bool forceUpdate);
 
+    angle::Result updateUniformBuffers(ContextMtl *context,
+                                       mtl::RenderCommandEncoder *cmdEncoder,
+                                       const mtl::RenderPipelineDesc &pipelineDesc);
+    angle::Result legalizeUniformBufferOffsets(ContextMtl *context,
+                                               const std::vector<gl::InterfaceBlock> &blocks);
+    angle::Result bindUniformBuffersToDiscreteSlots(ContextMtl *context,
+                                                    mtl::RenderCommandEncoder *cmdEncoder,
+                                                    const std::vector<gl::InterfaceBlock> &blocks,
+                                                    gl::ShaderType shaderType);
+    angle::Result encodeUniformBuffersInfoArgumentBuffer(
+        ContextMtl *context,
+        mtl::RenderCommandEncoder *cmdEncoder,
+        const std::vector<gl::InterfaceBlock> &blocks,
+        gl::ShaderType shaderType);
+
     void reset(ContextMtl *context);
+
     void linkResources(const gl::ProgramLinkedResources &resources);
     angle::Result linkImpl(const gl::Context *glContext,
                            const gl::ProgramLinkedResources &resources,
                            gl::InfoLog &infoLog);
 
-    angle::Result createMslShaderLib(const gl::Context *glContext,
+    angle::Result createMslShaderLib(mtl::Context *context,
                                      gl::ShaderType shaderType,
                                      gl::InfoLog &infoLog,
-                                     const std::string &translatedSource);
+                                     mtl::TranslatedShaderInfo *translatedMslInfo);
 
     // State for the default uniform blocks.
     struct DefaultUniformBlock final : private angle::NonCopyable
@@ -175,16 +207,25 @@ class ProgramMtl : public ProgramImpl, public mtl::RenderPipelineCacheSpecialize
     gl::ShaderBitSet mSamplerBindingsDirty;
     gl::ShaderMap<DefaultUniformBlock> mDefaultUniformBlocks;
 
-    gl::ShaderMap<std::string> mTranslatedMslShader;
-
+    // Translated metal shaders:
     gl::ShaderMap<mtl::TranslatedShaderInfo> mMslShaderTranslateInfo;
-    gl::ShaderMap<mtl::AutoObjCPtr<id<MTLLibrary>>> mMslShaderLibrary;
 
-    // Shader variants:
-    // - Vertex shader: One variant for now.
-    std::array<ProgramShaderVariantMtl, 1> mVertexShaderVariants;
+    // Compiled native shader object variants:
+    // - Vertex shader: one variant for now.
+    std::array<ProgramShaderObjVariantMtl, 1> mVertexShaderVariants;
     // - Fragment shader: One with sample coverage mask enabled, one with it disabled.
-    std::array<ProgramShaderVariantMtl, 2> mFragmentShaderVariants;
+    std::array<ProgramShaderObjVariantMtl, 2> mFragmentShaderVariants;
+
+    // Cached references of current shader variants.
+    gl::ShaderMap<ProgramShaderObjVariantMtl *> mCurrentShaderVariants;
+
+    // Scratch data:
+    // Legalized buffers and their offsets. For example, uniform buffer's offset=1 is not a valid
+    // offset, it will be converted to legal offset and the result is stored in this array.
+    std::vector<std::pair<mtl::BufferRef, uint32_t>> mLegalizedOffsetedUniformBuffers;
+    // Stores the render stages usage of each uniform buffer. Only used if the buffers are encoded
+    // into an argument buffer.
+    std::vector<uint32_t> mArgumentBufferRenderStageUsages;
 
     mtl::RenderPipelineCache mMetalRenderPipelineCache;
 };

src/libANGLE/renderer/metal/TransformFeedbackMtl.h

+//
+// Copyright 2020 The ANGLE Project Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+//
+// TransformFeedbackMtl.h:
+//    Defines the class interface for TransformFeedbackMtl, implementing TransformFeedbackImpl.
+//
+
+#ifndef LIBANGLE_RENDERER_METAL_TRANSFORMFEEDBACKMTL_H_
+#define LIBANGLE_RENDERER_METAL_TRANSFORMFEEDBACKMTL_H_
+
+#include "libANGLE/renderer/TransformFeedbackImpl.h"
+
+namespace gl
+{
+class ProgramState;
+}  // namespace gl
+
+namespace rx
+{
+
+class TransformFeedbackMtl : public TransformFeedbackImpl
+{
+  public:
+    TransformFeedbackMtl(const gl::TransformFeedbackState &state);
+    ~TransformFeedbackMtl() override;
+
+    angle::Result begin(const gl::Context *context, gl::PrimitiveMode primitiveMode) override;
+    angle::Result end(const gl::Context *context) override;
+    angle::Result pause(const gl::Context *context) override;
+    angle::Result resume(const gl::Context *context) override;
+
+    angle::Result bindIndexedBuffer(const gl::Context *context,
+                                    size_t index,
+                                    const gl::OffsetBindingPointer<gl::Buffer> &binding) override;
+
+  private:
+};
+
+}  // namespace rx
+
+#endif  // LIBANGLE_RENDERER_METAL_TRANSFORMFEEDBACKMTL_H_

src/libANGLE/renderer/metal/TransformFeedbackMtl.mm

+//
+// Copyright 2020 The ANGLE Project Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+//
+// TransformFeedbackMtl.mm:
+//    Defines the class interface for TransformFeedbackMtl, implementing TransformFeedbackImpl.
+//
+
+#include "libANGLE/renderer/metal/TransformFeedbackMtl.h"
+
+#include "libANGLE/Context.h"
+#include "libANGLE/Query.h"
+
+#include "common/debug.h"
+
+namespace rx
+{
+
+TransformFeedbackMtl::TransformFeedbackMtl(const gl::TransformFeedbackState &state)
+    : TransformFeedbackImpl(state)
+{}
+
+TransformFeedbackMtl::~TransformFeedbackMtl() {}
+
+angle::Result TransformFeedbackMtl::begin(const gl::Context *context,
+                                          gl::PrimitiveMode primitiveMode)
+{
+    UNIMPLEMENTED();
+
+    return angle::Result::Continue;
+}
+
+angle::Result TransformFeedbackMtl::end(const gl::Context *context)
+{
+    UNIMPLEMENTED();
+
+    return angle::Result::Continue;
+}
+
+angle::Result TransformFeedbackMtl::pause(const gl::Context *context)
+{
+    UNIMPLEMENTED();
+    return angle::Result::Continue;
+}
+
+angle::Result TransformFeedbackMtl::resume(const gl::Context *context)
+{
+    UNIMPLEMENTED();
+    return angle::Result::Continue;
+}
+
+angle::Result TransformFeedbackMtl::bindIndexedBuffer(
+    const gl::Context *context,
+    size_t index,
+    const gl::OffsetBindingPointer<gl::Buffer> &binding)
+{
+    UNIMPLEMENTED();
+
+    return angle::Result::Continue;
+}
+
+}  // namespace rx

src/libANGLE/renderer/metal/mtl_command_buffer.h

@@ -418,6 +418,10 @@ class RenderCommandEncoder final : public CommandEncoder
 
     RenderCommandEncoder &setVisibilityResultMode(MTLVisibilityResultMode mode, size_t offset);
 
+    RenderCommandEncoder &useResource(const BufferRef &resource,
+                                      MTLResourceUsage usage,
+                                      mtl::RenderStages states);
+
     RenderCommandEncoder &setColorStoreAction(MTLStoreAction action, uint32_t colorAttachmentIndex);
     // Set store action for every color attachment.
     RenderCommandEncoder &setColorStoreAction(MTLStoreAction action);

src/libANGLE/renderer/metal/mtl_command_buffer.mm

@@ -62,7 +62,8 @@ namespace
     PROC(DrawIndexed)                    \
     PROC(DrawIndexedInstanced)           \
     PROC(DrawIndexedInstancedBaseVertex) \
-    PROC(SetVisibilityResultMode)
+    PROC(SetVisibilityResultMode)        \
+    PROC(UseResource)
 
 #define ANGLE_MTL_TYPE_DECL(CMD) CMD,
 
@@ -325,6 +326,25 @@ void SetVisibilityResultModeCmd(id<MTLRenderCommandEncoder> encoder,
     [encoder setVisibilityResultMode:mode offset:offset];
 }
 
+void UseResourceCmd(id<MTLRenderCommandEncoder> encoder, IntermediateCommandStream *stream)
+{
+    id<MTLResource> resource = stream->fetch<id<MTLResource>>();
+    MTLResourceUsage usage   = stream->fetch<MTLResourceUsage>();
+    mtl::RenderStages stages = stream->fetch<mtl::RenderStages>();
+    ANGLE_UNUSED_VARIABLE(stages);
+#if defined(__IPHONE_13_0) || defined(__MAC_10_15)
+    if (ANGLE_APPLE_AVAILABLE_XCI(10.15, 13.0, 13.0))
+    {
+        [encoder useResource:resource usage:usage stages:stages];
+    }
+    else
+#endif
+    {
+        [encoder useResource:resource usage:usage];
+    }
+    [resource ANGLE_MTL_RELEASE];
+}
+
 // Command encoder mapping
 #define ANGLE_MTL_CMD_MAP(CMD) CMD##Cmd,
 
@@ -1422,6 +1442,25 @@ RenderCommandEncoder &RenderCommandEncoder::setVisibilityResultMode(MTLVisibilit
     return *this;
 }
 
+RenderCommandEncoder &RenderCommandEncoder::useResource(const BufferRef &resource,
+                                                        MTLResourceUsage usage,
+                                                        mtl::RenderStages states)
+{
+    if (!resource)
+    {
+        return *this;
+    }
+
+    cmdBuffer().setReadDependency(resource);
+
+    mCommands.push(CmdType::UseResource)
+        .push([resource->get() ANGLE_MTL_RETAIN])
+        .push(usage)
+        .push(states);
+
+    return *this;
+}
+
 RenderCommandEncoder &RenderCommandEncoder::setColorStoreAction(MTLStoreAction action,
                                                                 uint32_t colorAttachmentIndex)
 {

src/libANGLE/renderer/metal/mtl_common.h

@@ -99,6 +99,9 @@ constexpr uint32_t kMaxVertexAttribs = gl::MAX_VERTEX_ATTRIBS;
 // NOTE(hqle): support variable max number of render targets
 constexpr uint32_t kMaxRenderTargets = 4;
 
+constexpr uint32_t kMaxShaderUBOs = 12;
+constexpr uint32_t kMaxUBOSize    = 16384;
+
 // The max size of a buffer that will be allocated in shared memory.
 // NOTE(hqle): This is just a hint. There is no official document on what is the max allowed size
 // for shared memory.
@@ -120,10 +123,12 @@ constexpr uint32_t kUniformBufferSettingOffsetMinAlignment = 256;
 constexpr uint32_t kUniformBufferSettingOffsetMinAlignment = 4;
 #endif
 constexpr uint32_t kIndexBufferOffsetAlignment       = 4;
+constexpr uint32_t kArgumentBufferOffsetAlignment    = kUniformBufferSettingOffsetMinAlignment;
 constexpr uint32_t kTextureToBufferBlittingAlignment = 256;
 
 // Front end binding limits
 constexpr uint32_t kMaxGLSamplerBindings = 2 * kMaxShaderSamplers;
+constexpr uint32_t kMaxGLUBOBindings     = 2 * kMaxShaderUBOs;
 
 // Binding index start for vertex data buffers:
 constexpr uint32_t kVboBindingIndexStart = 0;
@@ -134,6 +139,8 @@ constexpr uint32_t kDefaultAttribsBindingIndex = kVboBindingIndexStart + kMaxVer
 constexpr uint32_t kDriverUniformsBindingIndex = kDefaultAttribsBindingIndex + 1;
 // Binding index for default uniforms:
 constexpr uint32_t kDefaultUniformsBindingIndex = kDefaultAttribsBindingIndex + 3;
+// Binding index for UBO's argument buffer or starting discrete slot
+constexpr uint32_t kUBOArgumentBufferBindingIndex = kDefaultUniformsBindingIndex + 1;
 
 constexpr uint32_t kStencilMaskAll = 0xff;  // Only 8 bits stencil is supported
 
@@ -146,8 +153,7 @@ constexpr float kEmulatedAlphaValue = 1.0f;
 
 constexpr size_t kOcclusionQueryResultSize = sizeof(uint64_t);
 
-// NOTE(hqle): Support ES 3.0.
-constexpr gl::Version kMaxSupportedGLVersion = gl::Version(2, 0);
+constexpr gl::Version kMaxSupportedGLVersion = gl::Version(3, 0);
 
 // Work-around the enum is not available on macOS
 #if TARGET_OS_OSX || TARGET_OS_MACCATALYST
@@ -156,6 +162,16 @@ constexpr MTLBlitOption kBlitOptionRowLinearPVRTC = MTLBlitOptionNone;
 constexpr MTLBlitOption kBlitOptionRowLinearPVRTC          = MTLBlitOptionRowLinearPVRTC;
 #endif
 
+#if defined(__IPHONE_13_0) || defined(__MAC_10_15)
+using RenderStages                             = MTLRenderStages;
+constexpr MTLRenderStages kRenderStageVertex   = MTLRenderStageVertex;
+constexpr MTLRenderStages kRenderStageFragment = MTLRenderStageFragment;
+#else
+using RenderStages                                         = int;
+constexpr RenderStages kRenderStageVertex                  = 1;
+constexpr RenderStages kRenderStageFragment                = 2;
+#endif
+
 enum class PixelType
 {
     Int,

src/libANGLE/renderer/metal/mtl_glslang_utils.h

@@ -28,8 +28,16 @@ struct SamplerBinding
 
 struct TranslatedShaderInfo
 {
+    void reset();
+
+    // Translated Metal source code
+    std::string metalShaderSource;
+    // Metal library compiled from source code above. Used by ProgramMtl.
+    AutoObjCPtr<id<MTLLibrary>> metalLibrary;
+
     std::array<SamplerBinding, kMaxGLSamplerBindings> actualSamplerBindings;
-    // NOTE(hqle): UBO, XFB bindings.
+    std::array<uint32_t, kMaxGLUBOBindings> actualUBOBindings;
+    bool hasUBOArgumentBuffer;
 };
 
 void GlslangGetShaderSource(const gl::ProgramState &programState,
@@ -48,8 +56,7 @@ angle::Result GlslangGetShaderSpirvCode(ErrorHandler *context,
 angle::Result SpirvCodeToMsl(Context *context,
                              const gl::ProgramState &programState,
                              gl::ShaderMap<std::vector<uint32_t>> *sprivShaderCode,
-                             gl::ShaderMap<TranslatedShaderInfo> *mslShaderInfoOut,
-                             gl::ShaderMap<std::string> *mslCodeOut);
+                             gl::ShaderMap<TranslatedShaderInfo> *mslShaderInfoOut);
 
 }  // namespace mtl
 }  // namespace rx

src/tests/gl_tests/UniformBufferTest.cpp

@@ -2403,7 +2403,7 @@ TEST_P(UniformBufferTest, UniformBlocksInDiffProgramShareUniformBuffer)
 // buffer data correctly.
 TEST_P(UniformBufferTest, UniformBlocksSharedSameUniformBuffer)
 {
-    ANGLE_SKIP_TEST_IF(IsIntel() && IsOSX());
+    ANGLE_SKIP_TEST_IF(IsIntel() && IsOSX() && IsOpenGL());
 
     GLint64 maxUniformBlockSize;
     glGetInteger64v(GL_MAX_UNIFORM_BLOCK_SIZE, &maxUniformBlockSize);
@@ -2842,7 +2842,7 @@ TEST_P(UniformBufferTest, UniformBlockWithOneLargeVec4Array)
 TEST_P(UniformBufferTest, UniformBlockWithOneLargeMatrixArrayWithRowMajorQualifier)
 {
     // http://anglebug.com/3837 , http://anglebug.com/2273
-    ANGLE_SKIP_TEST_IF(IsOSX() || IsAndroid() || (IsAMD() && IsOpenGL()) ||
+    ANGLE_SKIP_TEST_IF((IsOSX() && IsOpenGL()) || IsAndroid() || (IsAMD() && IsOpenGL()) ||
                        (IsLinux() && IsIntel() && IsOpenGL()));
     GLint64 maxUniformBlockSize;
     glGetInteger64v(GL_MAX_UNIFORM_BLOCK_SIZE, &maxUniformBlockSize);
@@ -3096,7 +3096,7 @@ TEST_P(UniformBufferTest, InstanceArrayUniformBlockWithOneLargeMatrixArray)
 TEST_P(UniformBufferTest, Std140UniformBlockWithDynamicallyIndexedRowMajorArray)
 {
     // http://anglebug.com/3837 , http://anglebug.com/2273
-    ANGLE_SKIP_TEST_IF((IsLinux() && IsIntel() && IsOpenGL()) || IsOSX());
+    ANGLE_SKIP_TEST_IF(((IsLinux() && IsIntel()) || IsOSX()) && IsOpenGL());
 
     constexpr char kFS[] =
         R"(#version 300 es
@@ -3180,9 +3180,78 @@ void main(void){
     EXPECT_GL_NO_ERROR();
 }
 
+// Test with many uniform buffers work as expected.
+TEST_P(UniformBufferTest, ManyBlocks)
+{
+    // http://anglebug.com/5039
+    ANGLE_SKIP_TEST_IF(IsD3D11());
+
+    constexpr char kFS[] =
+        R"(#version 300 es
+
+        precision highp float;
+        out vec4 my_FragColor;
+
+        layout(std140) uniform uboBlock { vec4 color; } blocks[12];
+
+        void main()
+        {
+            vec4 color = vec4(0, 0, 0, 1);
+            color += blocks[0].color;
+            color += blocks[1].color;
+            color += blocks[2].color;
+            color += blocks[3].color;
+            color += blocks[4].color;
+            color += blocks[5].color;
+            color += blocks[6].color;
+            color += blocks[7].color;
+            color += blocks[8].color;
+            color += blocks[9].color;
+            color += blocks[10].color;
+            color += blocks[11].color;
+            my_FragColor = vec4(color.rgb, 1.0);
+        })";
+
+    ANGLE_GL_PROGRAM(program, essl3_shaders::vs::Simple(), kFS);
+    GLBuffer buffers[12];
+    GLint bufferIndex[12];
+    bufferIndex[0]  = glGetUniformBlockIndex(program, "uboBlock[0]");
+    bufferIndex[1]  = glGetUniformBlockIndex(program, "uboBlock[1]");
+    bufferIndex[2]  = glGetUniformBlockIndex(program, "uboBlock[2]");
+    bufferIndex[3]  = glGetUniformBlockIndex(program, "uboBlock[3]");
+    bufferIndex[4]  = glGetUniformBlockIndex(program, "uboBlock[4]");
+    bufferIndex[5]  = glGetUniformBlockIndex(program, "uboBlock[5]");
+    bufferIndex[6]  = glGetUniformBlockIndex(program, "uboBlock[6]");
+    bufferIndex[7]  = glGetUniformBlockIndex(program, "uboBlock[7]");
+    bufferIndex[8]  = glGetUniformBlockIndex(program, "uboBlock[8]");
+    bufferIndex[9]  = glGetUniformBlockIndex(program, "uboBlock[9]");
+    bufferIndex[10] = glGetUniformBlockIndex(program, "uboBlock[10]");
+    bufferIndex[11] = glGetUniformBlockIndex(program, "uboBlock[11]");
+
+    std::vector<GLubyte> v(16, 0);
+    float *vAsFloat = reinterpret_cast<float *>(v.data());
+
+    for (int i = 0; i < 12; ++i)
+    {
+        glBindBuffer(GL_UNIFORM_BUFFER, buffers[i]);
+        vAsFloat[0] = (i + 1) / 255.0f;
+        vAsFloat[1] = (i + 1) / 255.0f;
+        vAsFloat[2] = (i + 1) / 255.0f;
+        vAsFloat[3] = 255.0f;
+
+        glBufferData(GL_UNIFORM_BUFFER, v.size(), v.data(), GL_STATIC_DRAW);
+
+        glBindBufferBase(GL_UNIFORM_BUFFER, i, buffers[i]);
+        glUniformBlockBinding(program, bufferIndex[i], i);
+    }
+
+    drawQuad(program.get(), essl3_shaders::PositionAttrib(), 0.5f);
+    EXPECT_PIXEL_NEAR(0, 0, 78, 78, 78, 255, 2);
+}
+
 // Use this to select which configurations (e.g. which renderer, which GLES major version) these
 // tests should be run against.
-ANGLE_INSTANTIATE_TEST_ES3(UniformBufferTest);
+ANGLE_INSTANTIATE_TEST_ES3_AND(UniformBufferTest, ES3_METAL());
 ANGLE_INSTANTIATE_TEST_ES31(UniformBufferTest31);
 
 }  // namespace