Initial -embed-spirv support (#4974)

* Initial -embed-spirv support

Add support for SPIR-V precompilation using the framework
established for DXIL.

Work on #4883

* SLANG_UNUSED

* Add linkage attributes to exported spirv functions

* Combine DXIL and SPIRV paths

* Whitespace fix

* Merge remaining precompiled spirv/dxil paths

* Change inst accessors to return codegentarget

* Add unit test for precompiled spirv

---------

Co-authored-by: Yong He <yonghe@outlook.com>

12 files changed, 161 insertions, 93 deletions

diff --git a/source/slang/slang-compiler-tu.cpp b/source/slang/slang-compiler-tu.cpp
index fe778148b..57a1133fc 100644
--- a/source/slang/slang-compiler-tu.cpp
+++ b/source/slang/slang-compiler-tu.cpp
@@ -54,7 +54,7 @@ namespace Slang
     * Precompile the module for the given target.
     *
     * This function creates a target program and emits the precompiled blob as
-    * an embedded blob in the module IR, e.g. DXIL.
+    * an embedded blob in the module IR, e.g. DXIL, SPIR-V.
     * Because the IR for the Slang Module may violate the restrictions of the
     * target language, the emitted target blob may not be able to include the
     * full module, but rather only the subset that can be precompiled. For
@@ -77,7 +77,7 @@ namespace Slang
     * in the linked IR which survived the additional pruning.
     *
     * Functions that are rejected after linking+legalization (inside
-    * emitPrecompiled*):
+    * emitPrecompiledDownstreamIR):
     * - (DXIL) Functions that return or take a HLSLStructuredBufferType
     * - (DXIL) Functions that return or take a Matrix type
     *
@@ -86,17 +86,13 @@ namespace Slang
     * phase of filtering.
     *
     * The original module IR functions matching those are then marked with
-    * "AvailableIn*" (e.g. AvailableInDXILDecoration) to indicate to future
+    * "AvailableInDownstreamIRDecoration" to indicate to future
     * module users which functions are present in the precompiled blob.
     */
     SLANG_NO_THROW SlangResult SLANG_MCALL Module::precompileForTarget(
         SlangCompileTarget target,
         slang::IBlob** outDiagnostics)
     {
-        if (target != SLANG_DXIL)
-        {
-            return SLANG_FAIL;
-        }
         CodeGenTarget targetEnum = CodeGenTarget(target);
 
         auto module = getIRModule();
@@ -130,10 +126,15 @@ namespace Slang
         {
         case CodeGenTarget::DXIL:
             tp.getOptionSet().add(CompilerOptionName::Profile, Profile::RawEnum::DX_Lib_6_6);
-            tp.getOptionSet().add(CompilerOptionName::EmbedDXIL, true);
             break;
+        case CodeGenTarget::SPIRV:
+            break;
+        default:
+            return SLANG_FAIL;
         }
 
+        tp.getOptionSet().add(CompilerOptionName::EmbedDownstreamIR, true);
+
         CodeGenContext::EntryPointIndices entryPointIndices;
 
         entryPointIndices.setCount(entryPointCount);
@@ -166,7 +167,7 @@ namespace Slang
         }
 
         ComPtr<IArtifact> outArtifact;
-        SlangResult res = codeGenContext.emitPrecompiledDXIL(outArtifact);
+        SlangResult res = codeGenContext.emitPrecompiledDownstreamIR(outArtifact);
 
         sink.getBlobIfNeeded(outDiagnostics);
         if (res != SLANG_OK)
@@ -183,7 +184,8 @@ namespace Slang
         for (const auto& mangledName : metadata->getExportedFunctionMangledNames())
         {
             auto moduleInst = nameToFunction[mangledName];
-            builder.addDecoration(moduleInst, kIROp_AvailableInDXILDecoration);
+            builder.addDecoration(moduleInst, kIROp_AvailableInDownstreamIRDecoration,
+                builder.getIntValue(builder.getIntType(), (int)targetReq->getTarget()));
             auto moduleDec = moduleInst->findDecoration<IRDownstreamModuleExportDecoration>();
             moduleDec->removeAndDeallocate();
         }
@@ -207,12 +209,7 @@ namespace Slang
         // Add the precompiled blob to the module
         builder.setInsertInto(module);
 
-        switch (targetReq->getTarget())
-        {
-        case CodeGenTarget::DXIL:
-            builder.emitEmbeddedDXIL(blob);
-            break;
-        }
+        builder.emitEmbeddedDownstreamIR(targetReq->getTarget(), blob);
 
         return SLANG_OK;
     }
diff --git a/source/slang/slang-compiler.cpp b/source/slang/slang-compiler.cpp
index 7d2dbed0d..4bb420fa7 100644
--- a/source/slang/slang-compiler.cpp
+++ b/source/slang/slang-compiler.cpp
@@ -767,9 +767,9 @@ namespace Slang
 #   pragma warning(pop)
 #endif
 
-    SlangResult CodeGenContext::emitPrecompiledDXIL(ComPtr<IArtifact>& outArtifact)
+    SlangResult CodeGenContext::emitPrecompiledDownstreamIR(ComPtr<IArtifact>& outArtifact)
     {
-        return emitWithDownstreamForEntryPoints(outArtifact);
+        return _emitEntryPoints(outArtifact);
     }
 
     String GetHLSLProfileName(Profile profile)
@@ -1257,10 +1257,7 @@ namespace Slang
         {
             CodeGenContext sourceCodeGenContext(this, sourceTarget, extensionTracker);
 
-            if (target == CodeGenTarget::DXILAssembly || target == CodeGenTarget::DXIL)
-            {
-                sourceCodeGenContext.removeAvailableInDXIL = true;
-            }
+            sourceCodeGenContext.removeAvailableInDownstreamIR = true;
 
             SLANG_RETURN_ON_FAIL(sourceCodeGenContext.emitEntryPointsSource(sourceArtifact));
             sourceCodeGenContext.maybeDumpIntermediate(sourceArtifact);
@@ -1561,20 +1558,23 @@ namespace Slang
         // Load embedded precompiled libraries from IR into library artifacts
         program->enumerateIRModules([&](IRModule* irModule)
         {
-            for (auto inst : irModule->getModuleInst()->getChildren())
+            for (auto globalInst : irModule->getModuleInst()->getChildren())
             {
                 if (target == CodeGenTarget::DXILAssembly || target == CodeGenTarget::DXIL)
                 {
-                    if (inst->getOp() == kIROp_EmbeddedDXIL)
+                    if (auto inst = as<IREmbeddedDownstreamIR>(globalInst))
                     {
-                        auto slice = static_cast<IRBlobLit*>(inst->getOperand(0))->getStringSlice();
-                        ArtifactDesc desc = ArtifactDescUtil::makeDescForCompileTarget(SLANG_DXIL);
-                        desc.kind = ArtifactKind::Library;
+                        if (inst->getTarget() == CodeGenTarget::DXIL)
+                        {
+                            auto slice = inst->getBlob()->getStringSlice();
+                            ArtifactDesc desc = ArtifactDescUtil::makeDescForCompileTarget(SLANG_DXIL);
+                            desc.kind = ArtifactKind::Library;
 
-                        auto library = ArtifactUtil::createArtifact(desc);
+                            auto library = ArtifactUtil::createArtifact(desc);
 
-                        library->addRepresentationUnknown(StringBlob::create(slice));
-                        libraries.add(library);
+                            library->addRepresentationUnknown(StringBlob::create(slice));
+                            libraries.add(library);
+                        }
                     }
                 }
             }
@@ -2030,7 +2030,7 @@ namespace Slang
             {                
                 if (auto artifact = targetProgram->getExistingWholeProgramResult())
                 {
-                    if (!targetProgram->getOptionSet().getBoolOption(CompilerOptionName::EmbedDXIL))
+                    if (!targetProgram->getOptionSet().getBoolOption(CompilerOptionName::EmbedDownstreamIR))
                     {
                         artifacts.add(ComPtr<IArtifact>(artifact));
                     }
@@ -2285,6 +2285,9 @@ namespace Slang
         auto linkage = getLinkage();
         for (auto targetReq : linkage->targets)
         {
+            if (targetReq->getOptionSet().getBoolOption(CompilerOptionName::EmbedDownstreamIR))
+                continue;
+
             auto targetProgram = program->getTargetProgram(targetReq);
             generateOutput(targetProgram);
         }
diff --git a/source/slang/slang-compiler.h b/source/slang/slang-compiler.h
index 267fdeaf5..b8ee4dc9c 100755
--- a/source/slang/slang-compiler.h
+++ b/source/slang/slang-compiler.h
@@ -2733,13 +2733,13 @@ namespace Slang
 
         SlangResult emitEntryPoints(ComPtr<IArtifact>& outArtifact);
 
-        SlangResult emitPrecompiledDXIL(ComPtr<IArtifact>& outArtifact);
+        SlangResult emitPrecompiledDownstreamIR(ComPtr<IArtifact>& outArtifact);
 
         void maybeDumpIntermediate(IArtifact* artifact);
 
-        // Used to cause instructions available in precompiled DXIL to be
+        // Used to cause instructions available in precompiled blobs to be
         // removed between IR linking and target source generation.
-        bool removeAvailableInDXIL = false;
+        bool removeAvailableInDownstreamIR = false;
 
     protected:
         CodeGenTarget m_targetFormat = CodeGenTarget::Unknown;
@@ -2816,7 +2816,7 @@ namespace Slang
         virtual SLANG_NO_THROW void SLANG_MCALL setTargetForceGLSLScalarBufferLayout(int targetIndex, bool value) SLANG_OVERRIDE;
         virtual SLANG_NO_THROW void SLANG_MCALL setTargetForceDXLayout(int targetIndex, bool value) SLANG_OVERRIDE;
         virtual SLANG_NO_THROW void SLANG_MCALL setTargetGenerateWholeProgram(int targetIndex, bool value) SLANG_OVERRIDE;
-        virtual SLANG_NO_THROW void SLANG_MCALL setEmbedDXIL(bool value) SLANG_OVERRIDE;
+        virtual SLANG_NO_THROW void SLANG_MCALL setTargetEmbedDownstreamIR(int targetIndex, bool value) SLANG_OVERRIDE;
         virtual SLANG_NO_THROW void SLANG_MCALL setMatrixLayoutMode(SlangMatrixLayoutMode mode) SLANG_OVERRIDE;
         virtual SLANG_NO_THROW void SLANG_MCALL setDebugInfoLevel(SlangDebugInfoLevel level) SLANG_OVERRIDE;
         virtual SLANG_NO_THROW void SLANG_MCALL setOptimizationLevel(SlangOptimizationLevel level) SLANG_OVERRIDE;
diff --git a/source/slang/slang-emit-spirv.cpp b/source/slang/slang-emit-spirv.cpp
index 9088e959a..a2da4801e 100644
--- a/source/slang/slang-emit-spirv.cpp
+++ b/source/slang/slang-emit-spirv.cpp
@@ -3933,6 +3933,17 @@ struct SPIRVEmitContext
             }
             break;
         }
+        case kIROp_DownstreamModuleExportDecoration:
+        {
+            requireSPIRVCapability(SpvCapabilityLinkage);
+            auto name = decoration->getParent()->findDecoration<IRExportDecoration>()->getMangledName();
+            emitInst(getSection(SpvLogicalSectionID::Annotations),
+                decoration,
+                SpvOpDecorate,
+                dstID,
+            SpvDecorationLinkageAttributes, name, SpvLinkageTypeExport);
+            break;
+        }
         // ...
         }
 
@@ -6847,12 +6858,27 @@ SlangResult emitSPIRVFromIR(
 #endif
 
     auto shouldPreserveParams = codeGenContext->getTargetProgram()->getOptionSet().getBoolOption(CompilerOptionName::PreserveParameters);
+    auto generateWholeProgram = codeGenContext->getTargetProgram()->getOptionSet().getBoolOption(CompilerOptionName::GenerateWholeProgram);
     for (auto inst : irModule->getGlobalInsts())
     {
         if (as<IRDebugSource>(inst))
+        {
             context.ensureInst(inst);
+        }
         if (shouldPreserveParams && as<IRGlobalParam>(inst))
+        {
             context.ensureInst(inst);
+        }
+        if (generateWholeProgram)
+        {
+            if (auto func = as<IRFunc>(inst))
+            {
+                if (func->findDecoration<IRDownstreamModuleExportDecoration>())
+                {
+                    context.ensureInst(inst);
+                }
+            }
+        }
     }
 
     // Emit source language info.
diff --git a/source/slang/slang-emit.cpp b/source/slang/slang-emit.cpp
index caa8ca8ea..ed9e90462 100644
--- a/source/slang/slang-emit.cpp
+++ b/source/slang/slang-emit.cpp
@@ -416,31 +416,47 @@ bool checkStaticAssert(IRInst* inst, DiagnosticSink* sink)
     return false;
 }
 
-static void unexportNonEmbeddableDXIL(IRModule* irModule)
+static void unexportNonEmbeddableIR(CodeGenTarget target, IRModule* irModule)
 {
     for (auto inst : irModule->getGlobalInsts())
     {
         if (inst->getOp() == kIROp_Func)
         {
-            // DXIL does not permit HLSLStructureBufferType in exported functions
-            // or sadly Matrices (https://github.com/shader-slang/slang/issues/4880)
-            auto type = as<IRFuncType>(inst->getFullType());
-            auto argCount = type->getOperandCount();
-            for (UInt aa = 0; aa < argCount; ++aa)
+            bool remove = false;
+            if (target == CodeGenTarget::HLSL)
             {
-                auto operand = type->getOperand(aa);
-                if (operand->getOp() == kIROp_HLSLStructuredBufferType ||
-                    operand->getOp() == kIROp_MatrixType)
+                // DXIL does not permit HLSLStructureBufferType in exported functions
+                // or sadly Matrices (https://github.com/shader-slang/slang/issues/4880)
+                auto type = as<IRFuncType>(inst->getFullType());
+                auto argCount = type->getOperandCount();
+                for (UInt aa = 0; aa < argCount; ++aa)
                 {
-                    if (auto dec = inst->findDecoration<IRPublicDecoration>())
+                    auto operand = type->getOperand(aa);
+                    if (operand->getOp() == kIROp_HLSLStructuredBufferType ||
+                        operand->getOp() == kIROp_MatrixType)
                     {
-                        dec->removeAndDeallocate();
+                        remove = true;
+                        break;
                     }
-                    if (auto dec = inst->findDecoration<IRDownstreamModuleExportDecoration>())
-                    {
-                        dec->removeAndDeallocate();
-                    }
-                    break;
+                }
+            }
+            else if (target == CodeGenTarget::SPIRV)
+            {
+                // SPIR-V does not allow exporting entry points
+                if (inst->findDecoration<IREntryPointDecoration>())
+                {
+                    remove = true;
+                }
+            }
+            if (remove)
+            {
+                if (auto dec = inst->findDecoration<IRPublicDecoration>())
+                {
+                    dec->removeAndDeallocate();
+                }
+                if (auto dec = inst->findDecoration<IRDownstreamModuleExportDecoration>())
+                {
+                    dec->removeAndDeallocate();
                 }
             }
         }
@@ -778,9 +794,9 @@ Result linkAndOptimizeIR(
         break;
     }
 
-    if (codeGenContext->removeAvailableInDXIL)
+    if (codeGenContext->removeAvailableInDownstreamIR)
     {
-        removeAvailableInDownstreamModuleDecorations(irModule);
+        removeAvailableInDownstreamModuleDecorations(target, irModule);
     }
 
     if (targetProgram->getOptionSet().shouldRunNonEssentialValidation())
@@ -1490,11 +1506,9 @@ Result linkAndOptimizeIR(
     auto metadata = new ArtifactPostEmitMetadata;
     outLinkedIR.metadata = metadata;
 
-    if (targetProgram->getOptionSet().getBoolOption(CompilerOptionName::EmbedDXIL))
+    if (targetProgram->getOptionSet().getBoolOption(CompilerOptionName::EmbedDownstreamIR))
     {
-        // We need to make sure that we don't try to export any functions that can't
-        // be part of a DXIL library interface, eg. with resources.
-        unexportNonEmbeddableDXIL(irModule);
+        unexportNonEmbeddableIR(target, irModule);
     }
 
     collectMetadata(irModule, *metadata);
diff --git a/source/slang/slang-ir-inst-defs.h b/source/slang/slang-ir-inst-defs.h
index 5afca48b3..afc09f480 100644
--- a/source/slang/slang-ir-inst-defs.h
+++ b/source/slang/slang-ir-inst-defs.h
@@ -806,7 +806,7 @@ INST_RANGE(BindingQuery, GetRegisterIndex, GetRegisterSpace)
     INST(NumThreadsDecoration,              numThreads,             3, 0)
     INST(WaveSizeDecoration,                waveSize,               1, 0)
 
-    INST(AvailableInDXILDecoration,         availableInDXIL,        0, 0)
+    INST(AvailableInDownstreamIRDecoration, availableInDownstreamIR, 1, 0)
 
         // Added to IRParam parameters to an entry point
     /* GeometryInputPrimitiveTypeDecoration */
@@ -1244,7 +1244,7 @@ INST(DebugVar, DebugVar, 4, 0)
 INST(DebugValue, DebugValue, 2, 0)
 
 /* Embedded Precompiled Libraries */
-INST(EmbeddedDXIL, EmbeddedDXIL, 1, 0)
+INST(EmbeddedDownstreamIR, EmbeddedDownstreamIR, 2, 0)
 
 /* Inline assembly */
 
diff --git a/source/slang/slang-ir-insts.h b/source/slang/slang-ir-insts.h
index fc963697a..a0ed8ff0e 100644
--- a/source/slang/slang-ir-insts.h
+++ b/source/slang/slang-ir-insts.h
@@ -432,9 +432,14 @@ IR_SIMPLE_DECORATION(NonCopyableTypeDecoration)
 IR_SIMPLE_DECORATION(HLSLMeshPayloadDecoration)
 IR_SIMPLE_DECORATION(GlobalInputDecoration)
 IR_SIMPLE_DECORATION(GlobalOutputDecoration)
-IR_SIMPLE_DECORATION(AvailableInDXILDecoration)
 IR_SIMPLE_DECORATION(DownstreamModuleExportDecoration)
 
+struct IRAvailableInDownstreamIRDecoration : IRDecoration
+{
+    IR_LEAF_ISA(AvailableInDownstreamIRDecoration)
+    CodeGenTarget getTarget() { return static_cast<CodeGenTarget>(cast<IRIntLit>(getOperand(0))->getValue()); }
+};
+
 struct IRGLSLLocationDecoration : IRDecoration
 {
     IR_LEAF_ISA(GLSLLocationDecoration)
@@ -3340,6 +3345,13 @@ struct IRStaticAssert : IRInst
     IR_LEAF_ISA(StaticAssert)
 };
 
+struct IREmbeddedDownstreamIR : IRInst
+{
+    IR_LEAF_ISA(EmbeddedDownstreamIR)
+    CodeGenTarget getTarget() { return static_cast<CodeGenTarget>(cast<IRIntLit>(getOperand(0))->getValue()); }
+    IRBlobLit* getBlob() { return cast<IRBlobLit>(getOperand(1)); }
+};
+
 struct IRBuilderSourceLocRAII;
 
 struct IRBuilder
@@ -4053,7 +4065,7 @@ public:
     IRInst* emitByteAddressBufferStore(IRInst* byteAddressBuffer, IRInst* offset, IRInst* value);
     IRInst* emitByteAddressBufferStore(IRInst* byteAddressBuffer, IRInst* offset, IRInst* alignment, IRInst* value);
 
-    IRInst* emitEmbeddedDXIL(ISlangBlob* blob);
+    IRInst* emitEmbeddedDownstreamIR(CodeGenTarget target, ISlangBlob* blob);
 
     IRFunc* createFunc();
     IRGlobalVar* createGlobalVar(
diff --git a/source/slang/slang-ir-redundancy-removal.cpp b/source/slang/slang-ir-redundancy-removal.cpp
index 038412fbf..79b88ebcf 100644
--- a/source/slang/slang-ir-redundancy-removal.cpp
+++ b/source/slang/slang-ir-redundancy-removal.cpp
@@ -159,26 +159,29 @@ bool removeRedundancyInFunc(IRGlobalValueWithCode* func)
     return result;
 }
 
-// Remove IR definitions from all [AvailableInDXIL] functions when compiling DXIL,
-// as these functions are already defined in the embedded precompiled DXIL library.
-void removeAvailableInDownstreamModuleDecorations(IRModule* module)
+// Remove IR definitions from all AvailableInDownstreamIR functions where the
+// languages match what we're currently targetting,  as these functions are
+// already defined in the embedded precompiled library.
+void removeAvailableInDownstreamModuleDecorations(CodeGenTarget target, IRModule* module)
 {
     List<IRInst*> toRemove;
     for (auto globalInst : module->getGlobalInsts())
     {
-        auto funcInst = as<IRFunc>(globalInst);
-        if (!funcInst)
+        if (auto funcInst = as<IRFunc>(globalInst))
         {
-            continue;
-        }
-        if (globalInst->findDecoration<IRAvailableInDXILDecoration>())
-        {
-            // Gut the function definition, turning it into a declaration
-            for (auto inst : funcInst->getChildren())
+            if (auto dec = globalInst->findDecoration<IRAvailableInDownstreamIRDecoration>())
             {
-                if (inst->getOp() == kIROp_Block)
+                if ((dec->getTarget() == CodeGenTarget::DXIL && target == CodeGenTarget::HLSL) ||
+                    (dec->getTarget() == target))
                 {
-                    toRemove.add(inst);
+                    // Gut the function definition, turning it into a declaration
+                    for (auto inst : funcInst->getChildren())
+                    {
+                        if (inst->getOp() == kIROp_Block)
+                        {
+                            toRemove.add(inst);
+                        }
+                    }
                 }
             }
         }
diff --git a/source/slang/slang-ir-redundancy-removal.h b/source/slang/slang-ir-redundancy-removal.h
index 117f52708..8b006c683 100644
--- a/source/slang/slang-ir-redundancy-removal.h
+++ b/source/slang/slang-ir-redundancy-removal.h
@@ -1,5 +1,6 @@
 // slang-ir-redundancy-removal.h
 #pragma once
+#include "slang-compiler.h"
 
 namespace Slang
 {
@@ -11,5 +12,5 @@ namespace Slang
 
     bool eliminateRedundantLoadStore(IRGlobalValueWithCode* func);
 
-    void removeAvailableInDownstreamModuleDecorations(IRModule* module);
+    void removeAvailableInDownstreamModuleDecorations(CodeGenTarget target, IRModule* module);
 }
diff --git a/source/slang/slang-ir.cpp b/source/slang/slang-ir.cpp
index d02c01105..9305d1783 100644
--- a/source/slang/slang-ir.cpp
+++ b/source/slang/slang-ir.cpp
@@ -3863,11 +3863,11 @@ namespace Slang
         return nullptr;
     }
 
-    IRInst* IRBuilder::emitEmbeddedDXIL(ISlangBlob *blob)
+    IRInst* IRBuilder::emitEmbeddedDownstreamIR(CodeGenTarget target, ISlangBlob *blob)
     {
-        IRInst* args[] = { getBlobValue(blob) };
+        IRInst* args[] = { getIntValue(getIntType(), (int)target), getBlobValue(blob) };
 
-        return emitIntrinsicInst(getVoidType(), kIROp_EmbeddedDXIL, 1, args);
+        return emitIntrinsicInst(getVoidType(), kIROp_EmbeddedDownstreamIR, 2, args);
     }
 
     enum class TypeCastStyle
diff --git a/source/slang/slang-options.cpp b/source/slang/slang-options.cpp
index f12a66790..c02a00957 100644
--- a/source/slang/slang-options.cpp
+++ b/source/slang/slang-options.cpp
@@ -354,8 +354,6 @@ void initCommandOptions(CommandOptions& options)
         { OptionKind::DisableShortCircuit, "-disable-short-circuit", nullptr, "Disable short-circuiting for \"&&\" and \"||\" operations" },
         { OptionKind::UnscopedEnum, "-unscoped-enum", nullptr, "Treat enums types as unscoped by default."},
         { OptionKind::PreserveParameters, "-preserve-params", nullptr, "Preserve all resource parameters in the output code, even if they are not used by the shader."},
-        { OptionKind::EmbedDXIL, "-embed-dxil", nullptr,
-        "Embed DXIL into emitted slang-modules for faster linking" },
     };
 
     _addOptions(makeConstArrayView(generalOpts), options);
@@ -431,6 +429,7 @@ void initCommandOptions(CommandOptions& options)
         "A path to a specific spirv.core.grammar.json to use when generating SPIR-V output" },
         { OptionKind::IncompleteLibrary, "-incomplete-library", nullptr,
         "Allow generating code from incomplete libraries with unresolved external functions" },
+        { OptionKind::EmbedDownstreamIR, "-embed-downstream-ir", nullptr, "Embed downstream IR into emitted slang IR" },
     };
 
     _addOptions(makeConstArrayView(targetOpts), options);
@@ -1931,7 +1930,11 @@ SlangResult OptionsParser::_parse(
                 linkage->m_optionSet.set(optionKind, compressionType);
                 break;
             }
-            case OptionKind::EmbedDXIL: m_compileRequest->setEmbedDXIL(true); break;
+            case OptionKind::EmbedDownstreamIR:
+            {
+                getCurrentTarget()->optionSet.add(CompilerOptionName::EmbedDownstreamIR, true);
+                break;
+            }
             case OptionKind::Target:
             {
                 CommandLineArg name;
@@ -2778,6 +2781,12 @@ SlangResult OptionsParser::_parse(
             {
                 m_compileRequest->setTargetGenerateWholeProgram(targetID, true);
             }
+
+            if (rawTarget.optionSet.getBoolOption(CompilerOptionName::EmbedDownstreamIR))
+            {
+                m_compileRequest->setTargetEmbedDownstreamIR(targetID, true);
+            }
+
         }
 
         // Next we need to sort out the output files specified with `-o`, and
diff --git a/source/slang/slang.cpp b/source/slang/slang.cpp
index b0898b1a0..91ed3de5f 100644
--- a/source/slang/slang.cpp
+++ b/source/slang/slang.cpp
@@ -3305,18 +3305,21 @@ SlangResult EndToEndCompileRequest::executeActionsInner()
         return SLANG_OK;
     }
 
-    // If requested, attempt to compile the translation unit all the way down to the target language
-    // and stash the result blob in an IR op.
-    if (getOptionSet().getBoolOption(CompilerOptionName::EmbedDXIL))
+    // If requested, attempt to compile the translation unit all the way down to the target language(s)
+    // and stash the result blobs in IR.
+    for (auto target : getLinkage()->targets)
     {
-        auto frontEndReq = getFrontEndReq();
-
-        for (auto translationUnit : frontEndReq->translationUnits)
+        SlangCompileTarget targetEnum = SlangCompileTarget(target->getTarget());
+        if (target->getOptionSet().getBoolOption(CompilerOptionName::EmbedDownstreamIR))
         {
-            SlangCompileTarget target = SlangCompileTarget(SlangCompileTarget::SLANG_DXIL);
-            SLANG_RETURN_ON_FAIL(translationUnit->getModule()->precompileForTarget(
-                target,
-                nullptr));
+            auto frontEndReq = getFrontEndReq();
+
+            for (auto translationUnit : frontEndReq->translationUnits)
+            {
+                SLANG_RETURN_ON_FAIL(translationUnit->getModule()->precompileForTarget(
+                    targetEnum,
+                    nullptr));
+            }
         }
     }
 
@@ -5939,9 +5942,9 @@ void EndToEndCompileRequest::setTargetGenerateWholeProgram(int targetIndex, bool
     getTargetOptionSet(targetIndex).set(CompilerOptionName::GenerateWholeProgram, value);
 }
 
-void EndToEndCompileRequest::setEmbedDXIL(bool value)
+void EndToEndCompileRequest::setTargetEmbedDownstreamIR(int targetIndex, bool value)
 {
-    getOptionSet().set(CompilerOptionName::EmbedDXIL, value);
+    getTargetOptionSet(targetIndex).set(CompilerOptionName::EmbedDownstreamIR, value);
 }
 
 void EndToEndCompileRequest::setTargetLineDirectiveMode(