Use symbol alias instead of wrapper synthesis to implement link-time types. (#8603)

This change achieves link-time type resolution with a different
mechanism.

For `extern struct Foo : IFoo = FooImpl;`,
instead of synthesizing a wrapper type `Foo` that has a `FooImpl inner`
field and dispatches all interface method calls to `inner.method()`,
this PR completely removes this synthesis step, and instead just lower
such `extern`/`export` types as `IRSymbolAlias` instructions that is
just a reference to the type being wrapped.

Then we extend the linker logic to clone the referenced symbol instead
of the SymbolAlias insts itself during linking.

By doing so, we greatly simply the logic need to support link-time
types, and achieves higher robustness by not having to deal with many
AST synthesis scenarios.

Closes #8554.

---------

Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>

1 files changed, 229 insertions, 0 deletions

diff --git a/tools/gfx-unit-test/link-time-type-generic.cpp b/tools/gfx-unit-test/link-time-type-generic.cpp
new file mode 100644
index 000000000..ac1750518
--- /dev/null
+++ b/tools/gfx-unit-test/link-time-type-generic.cpp
@@ -0,0 +1,229 @@
+#include "core/slang-basic.h"
+#include "core/slang-blob.h"
+#include "gfx-test-util.h"
+#include "slang-rhi.h"
+#include "slang-rhi/shader-cursor.h"
+#include "unit-test/slang-unit-test.h"
+
+using namespace rhi;
+
+// Test that generic link time types conforming to a generic interface with generic
+// methods/subscript members work correctly.
+// Also test that global generic link-time functions works correctly.
+
+namespace gfx_test
+{
+static Slang::Result loadProgram(
+    rhi::IDevice* device,
+    Slang::ComPtr<rhi::IShaderProgram>& outShaderProgram,
+    slang::ProgramLayout*& slangReflection,
+    bool linkSpecialization = false)
+{
+    const char* moduleInterfaceSrc = R"(
+            interface ISimple { float getVal(); }
+            interface IHasProperty { property float val2{get;set;} }
+            interface IFoo<T:__BuiltinFloatingPointType> : IHasProperty
+            {
+                static const int offset;
+                [mutating] void setValue(float v);
+
+                T getValue<U:ISimple>(U u);
+
+                __subscript<U:__BuiltinIntegerType>(U index) -> T { get; }
+            }
+            struct FooImpl<T:__BuiltinFloatingPointType, int x> : IFoo<T>
+            {
+                T val;
+                static const int offset = x;
+                [mutating] void setValue(float v) { val = T(v); }
+                T getValue<U:ISimple>(U u){ return val + T(u.getVal()); }
+                property float val2 {
+                    get { return __real_cast<float>(val) + 2.0; }
+                    set { val = T(newValue); }
+                }
+                __subscript<U:__BuiltinIntegerType>(U index) -> T { get {return T(1.0); } }
+            };
+            struct BarImpl<T:__BuiltinFloatingPointType, int x> : IFoo<T>
+            {
+                T val;
+                static const int offset = -x;
+                [mutating] void setValue(float v) { val = T(v); }
+                T getValue<U:ISimple>(U u){ return val - T(1.0); }
+                property float val2 {
+                    get { return __real_cast<float>(val) + 2.0; }
+                    set { val = T(newValue); }
+                }
+                __subscript<U:__BuiltinIntegerType>(U index) -> T { get {return T(2.0); } }
+            };
+        )";
+    const char* module0Src = R"(
+            import ifoo;
+            extern struct Foo<T:__BuiltinFloatingPointType, int i> : IFoo<T> = FooImpl<T, i+1>;
+            extern static const float c = 0.0;
+            extern int linkTimeFunc<int x>() { return x; }
+            struct SimpleImpl : ISimple
+            {
+                float getVal() { return 100.0; }
+            };
+
+            // Use an indirect generic function to retrieve val2, to make sure intermediate witness tables
+            // can be obtained correctly from link-time witnesses.
+            float getVal2<T:IHasProperty>(T t) { return t.val2; }
+
+            [numthreads(1,1,1)]
+            void computeMain(uniform RWStructuredBuffer<float> buffer)
+            {
+                Foo<float, 0> foo;
+                foo.setValue(3.0);
+                buffer[0] = foo.getValue(SimpleImpl()) + getVal2(foo) + Foo<float, 0>.offset + c + foo[0] + linkTimeFunc<0>();
+            }
+        )";
+    const char* module1Src = R"(
+            import ifoo;
+            export struct Foo<T1:__BuiltinFloatingPointType, int i> : IFoo<T1> = BarImpl<T1, i+1>;
+            export static const float c = 1.0;
+            export int linkTimeFunc<int x>() { return x + 1; }
+        )";
+    Slang::ComPtr<slang::ISession> slangSession;
+    SLANG_RETURN_ON_FAIL(device->getSlangSession(slangSession.writeRef()));
+    Slang::ComPtr<slang::IBlob> diagnosticsBlob;
+    auto moduleInterfaceBlob =
+        Slang::UnownedRawBlob::create(moduleInterfaceSrc, strlen(moduleInterfaceSrc));
+    auto module0Blob = Slang::UnownedRawBlob::create(module0Src, strlen(module0Src));
+    auto module1Blob = Slang::UnownedRawBlob::create(module1Src, strlen(module1Src));
+    slang::IModule* moduleInterface =
+        slangSession->loadModuleFromSource("ifoo", "ifoo.slang", moduleInterfaceBlob);
+    slang::IModule* module0 = slangSession->loadModuleFromSource("module0", "path0", module0Blob);
+    slang::IModule* module1 = slangSession->loadModuleFromSource("module1", "path1", module1Blob);
+    ComPtr<slang::IEntryPoint> computeEntryPoint;
+    SLANG_RETURN_ON_FAIL(
+        module0->findEntryPointByName("computeMain", computeEntryPoint.writeRef()));
+
+    Slang::List<slang::IComponentType*> componentTypes;
+    componentTypes.add(moduleInterface);
+    componentTypes.add(module0);
+    if (linkSpecialization)
+        componentTypes.add(module1);
+    componentTypes.add(computeEntryPoint);
+
+    Slang::ComPtr<slang::IComponentType> composedProgram;
+    SlangResult result = slangSession->createCompositeComponentType(
+        componentTypes.getBuffer(),
+        componentTypes.getCount(),
+        composedProgram.writeRef(),
+        diagnosticsBlob.writeRef());
+    diagnoseIfNeeded(diagnosticsBlob);
+    SLANG_RETURN_ON_FAIL(result);
+
+    ComPtr<slang::IComponentType> linkedProgram;
+    result = composedProgram->link(linkedProgram.writeRef(), diagnosticsBlob.writeRef());
+    diagnoseIfNeeded(diagnosticsBlob);
+    SLANG_RETURN_ON_FAIL(result);
+
+    composedProgram = linkedProgram;
+    slangReflection = composedProgram->getLayout();
+
+    ShaderProgramDesc programDesc = {};
+    programDesc.slangGlobalScope = composedProgram.get();
+
+    auto shaderProgram = device->createShaderProgram(programDesc);
+
+    outShaderProgram = shaderProgram;
+    return SLANG_OK;
+}
+
+void linkTimeTypeGenericTestImpl(IDevice* device, UnitTestContext* context)
+{
+    // Create pipeline without linking a specialization override module, so we should
+    // see the default value of `extern Foo`.
+    ComPtr<IShaderProgram> shaderProgram;
+    slang::ProgramLayout* slangReflection;
+    GFX_CHECK_CALL_ABORT(loadProgram(device, shaderProgram, slangReflection, false));
+
+    ComputePipelineDesc pipelineDesc = {};
+    pipelineDesc.program = shaderProgram.get();
+    ComPtr<IComputePipeline> pipelineState;
+    GFX_CHECK_CALL_ABORT(device->createComputePipeline(pipelineDesc, pipelineState.writeRef()));
+
+    // Create pipeline with a specialization override module linked in, so we should
+    // see the result of using `BarImpl<T>` for `extern Foo<T>`.
+    ComPtr<IShaderProgram> shaderProgram1;
+    GFX_CHECK_CALL_ABORT(loadProgram(device, shaderProgram1, slangReflection, true));
+
+    ComputePipelineDesc pipelineDesc1 = {};
+    pipelineDesc1.program = shaderProgram1.get();
+    ComPtr<IComputePipeline> pipelineState1;
+    GFX_CHECK_CALL_ABORT(device->createComputePipeline(pipelineDesc1, pipelineState1.writeRef()));
+
+    const int numberCount = 4;
+    float initialData[] = {0.0f, 0.0f, 0.0f, 0.0f};
+    BufferDesc bufferDesc = {};
+    bufferDesc.size = numberCount * sizeof(float);
+    bufferDesc.format = rhi::Format::Undefined;
+    bufferDesc.elementSize = sizeof(float);
+    bufferDesc.usage = BufferUsage::ShaderResource | BufferUsage::UnorderedAccess |
+                       BufferUsage::CopyDestination | BufferUsage::CopySource;
+    bufferDesc.defaultState = ResourceState::UnorderedAccess;
+    bufferDesc.memoryType = MemoryType::DeviceLocal;
+
+    ComPtr<IBuffer> numbersBuffer;
+    GFX_CHECK_CALL_ABORT(
+        device->createBuffer(bufferDesc, (void*)initialData, numbersBuffer.writeRef()));
+
+    auto queue = device->getQueue(QueueType::Graphics);
+
+    // We have done all the set up work, now it is time to start recording a command buffer for
+    // GPU execution.
+    {
+        auto commandEncoder = queue->createCommandEncoder();
+        auto computePassEncoder = commandEncoder->beginComputePass();
+
+        auto rootObject = computePassEncoder->bindPipeline(pipelineState);
+
+        ShaderCursor entryPointCursor(
+            rootObject->getEntryPoint(0)); // get a cursor the the first entry-point.
+        // Bind buffer to the entry point.
+        entryPointCursor.getPath("buffer").setBinding(Binding(numbersBuffer));
+
+        computePassEncoder->dispatchCompute(1, 1, 1);
+        computePassEncoder->end();
+        auto commandBuffer = commandEncoder->finish();
+        queue->submit(commandBuffer);
+        queue->waitOnHost();
+    }
+
+    compareComputeResult(device, numbersBuffer, std::array{110.0f});
+
+    // Now run again with the overrided program.
+    {
+        auto commandEncoder = queue->createCommandEncoder();
+        auto computePassEncoder = commandEncoder->beginComputePass();
+
+        auto rootObject = computePassEncoder->bindPipeline(pipelineState1);
+
+        ShaderCursor entryPointCursor(
+            rootObject->getEntryPoint(0)); // get a cursor the the first entry-point.
+        // Bind buffer to the entry point.
+        entryPointCursor.getPath("buffer").setBinding(Binding(numbersBuffer));
+
+        computePassEncoder->dispatchCompute(1, 1, 1);
+        computePassEncoder->end();
+        auto commandBuffer = commandEncoder->finish();
+        queue->submit(commandBuffer);
+        queue->waitOnHost();
+    }
+
+    compareComputeResult(device, numbersBuffer, std::array{10.0f});
+}
+
+SLANG_UNIT_TEST(linkTimeTypeGenericD3D12)
+{
+    runTestImpl(linkTimeTypeGenericTestImpl, unitTestContext, DeviceType::D3D12);
+}
+
+SLANG_UNIT_TEST(linkTimeTypeGenerictVulkan)
+{
+    runTestImpl(linkTimeTypeGenericTestImpl, unitTestContext, DeviceType::Vulkan);
+}
+
+} // namespace gfx_test