[DirectX] DXILBitcodeWriter is trying to encode unsupported TargetExtTyID

[farzonl]

After #151751 merges These bugs that were origionally Type asserts because we were trying to cast Loads to CallInst become Bytecode encoding bugs
Ie before it was:

Assertion: (isa<To>(Val) && "cast<Ty>() argument of incompatible type!")
Function: cast
File: Casting.h:578

And now it is llvm_unreachable("These should never be used!!!"); Because we are trying to encode the TargetExtTyID.

#151751 is improving things, by fixing the DXIL ForwardHandle pass, but that is exposing some changes needed in the bitcode writer. Which we will handle as a seperate issue. Until this change merges this issue is not exposed even though it exists.

(lldb) frame select 4
frame #4: 0x0000000105433288 clang-dxc`llvm::dxil::DXILBitcodeWriter::writeTypeTable(this=0x000000016fdf3d98) at DXILBitcodeWriter.cpp:1025:7
   1022     case Type::X86_AMXTyID:
   1023     case Type::TokenTyID:
   1024     case Type::TargetExtTyID:
-> 1025       llvm_unreachable("These should never be used!!!");
   1026       break;
   1027     case Type::VoidTyID:
   1028       Code = bitc::TYPE_CODE_VOID;
(lldb) expr T->dump()
target("dx.RawBuffer", i32, 1, 0)

Crash Stack

(lldb) bt
* thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGABRT
  * frame #0: 0x000000018cb79388 libsystem_kernel.dylib`__pthread_kill + 8
    frame #1: 0x000000018cbb288c libsystem_pthread.dylib`pthread_kill + 296
    frame #2: 0x000000018cabbc60 libsystem_c.dylib`abort + 124
    frame #3: 0x000000010319bc10 clang-dxc`llvm::llvm_unreachable_internal(msg="These should never be used!!!", file="/Users/farzonlotfi/Projects/llvm-project/llvm/lib/Target/DirectX/DXILWriter/DXILBitcodeWriter.cpp", line=1025) at ErrorHandling.cpp:244:3
    frame #4: 0x0000000105433288 clang-dxc`llvm::dxil::DXILBitcodeWriter::writeTypeTable(this=0x000000016fdf3d98) at DXILBitcodeWriter.cpp:1025:7
    frame #5: 0x0000000105430abc clang-dxc`llvm::dxil::DXILBitcodeWriter::write(this=0x000000016fdf3d98) at DXILBitcodeWriter.cpp:2888:3
    frame #6: 0x0000000105430780 clang-dxc`llvm::dxil::BitcodeWriter::writeModule(this=0x000000016fdf40f0, M=0x0000000126e0f6c0) at DXILBitcodeWriter.cpp:436:16
    frame #7: 0x0000000105430628 clang-dxc`llvm::dxil::WriteDXILToFile(M=0x0000000126e0f6c0, Out=0x000000016fdf43c0) at DXILBitcodeWriter.cpp:406:10
    frame #8: 0x000000010546ab18 clang-dxc`(anonymous namespace)::EmbedDXILPass::runOnModule(this=0x000000011ca1ce40, M=0x0000000126e0f6c0) at DXILWriterPass.cpp:125:5
    frame #9: 0x0000000102003bd0 clang-dxc`(anonymous namespace)::MPPassManager::runOnModule(this=0x000000011ca1d850, M=0x0000000126e0f6c0) at LegacyPassManager.cpp:1513:27
    frame #10: 0x0000000102003754 clang-dxc`llvm::legacy::PassManagerImpl::run(this=0x000000011e02e200, M=0x0000000126e0f6c0) at LegacyPassManager.cpp:531:44
    frame #11: 0x000000010200a2ac clang-dxc`llvm::legacy::PassManager::run(this=0x000000016fdf47f8, M=0x0000000126e0f6c0) at LegacyPassManager.cpp:1640:14
    frame #12: 0x0000000103c67294 clang-dxc`(anonymous namespace)::EmitAssemblyHelper::RunCodegenPipeline(this=0x000000016fdf4c70, Action=Backend_EmitObj, OS=llvm::raw_pwrite_stream @ 0x0000000126e0f340, DwoOS=nullptr) at BackendUtil.cpp:1254:19
    frame #13: 0x0000000103c57a34 clang-dxc`(anonymous namespace)::EmitAssemblyHelper::emitAssembly(this=0x000000016fdf4c70, Action=Backend_EmitObj, OS=llvm::raw_pwrite_stream @ 0x0000000126e0f340, BC=0x0000000126e0f490) at BackendUtil.cpp:1278:3
    frame #14: 0x0000000103c56f28 clang-dxc`clang::emitBackendOutput(CI=0x0000000126f29060, CGOpts=0x000000012781e018, TDesc=(Data = "e-m:e-p:32:32-i1:32-i8:8-i16:16-i32:32-i64:64-f16:16-f32:32-f64:64-n8:16:32:64", Length = 78), M=0x0000000126e0f6c0, Action=Backend_EmitObj, VFS=IntrusiveRefCntPtr<llvm::vfs::FileSystem> @ 0x000000016fdf4f78, OS=nullptr, BC=0x0000000126e0f490) at BackendUtil.cpp:1443:13
    frame #15: 0x00000001043e9244 clang-dxc`clang::BackendConsumer::HandleTranslationUnit(this=0x0000000126e0f490, C=0x000000012701ce00) at CodeGenAction.cpp:316:3
    frame #16: 0x0000000106ca45d4 clang-dxc`clang::ParseAST(S=0x000000012703f000, PrintStats=false, SkipFunctionBodies=false) at ParseAST.cpp:183:13
    frame #17: 0x0000000104f9e2f8 clang-dxc`clang::ASTFrontendAction::ExecuteAction(this=0x0000000126f2afc0) at FrontendAction.cpp:1342:3
    frame #18: 0x00000001043eed68 clang-dxc`clang::CodeGenAction::ExecuteAction(this=0x0000000126f2afc0) at CodeGenAction.cpp:1113:30
    frame #19: 0x0000000104f9db68 clang-dxc`clang::FrontendAction::Execute(this=0x0000000126f2afc0) at FrontendAction.cpp:1222:3
    frame #20: 0x0000000104eb452c clang-dxc`clang::CompilerInstance::ExecuteAction(this=0x0000000126f29060, Act=0x0000000126f2afc0) at CompilerInstance.cpp:1061:33
    frame #21: 0x00000001050e1be8 clang-dxc`clang::ExecuteCompilerInvocation(Clang=0x0000000126f29060) at ExecuteCompilerInvocation.cpp:299:25
    frame #22: 0x0000000100012b98 clang-dxc`cc1_main(Argv=ArrayRef<const char *> @ 0x000000016fdf5e78, Argv0="/Users/farzonlotfi/Projects/llvm_debug_build/bin/clang-20", MainAddr=0x0000000100003684) at cc1_main.cpp:297:15
    frame #23: 0x0000000100005018 clang-dxc`ExecuteCC1Tool(ArgV=0x000000016fdf9758, ToolContext=0x000000016fdfe6f8) at driver.cpp:223:12
    frame #24: 0x0000000100011590 clang-dxc`clang_main(int, char**, llvm::ToolContext const&)::$_0::operator()(this=0x000000016fdfa290, ArgV=0x000000016fdf9758) const at driver.cpp:368:16
    frame #25: 0x0000000100011560 clang-dxc`int llvm::function_ref<int (llvm::SmallVectorImpl<char const*>&)>::callback_fn<clang_main(int, char**, llvm::ToolContext const&)::$_0>(callable=6171894416, params=0x000000016fdf9758) at STLFunctionalExtras.h:46:12
    frame #26: 0x0000000104bb1ea8 clang-dxc`llvm::function_ref<int (llvm::SmallVectorImpl<char const*>&)>::operator()(this=0x000000016fdfa918, params=0x000000016fdf9758) const at STLFunctionalExtras.h:69:12
    frame #27: 0x0000000104bb1e68 clang-dxc`clang::driver::CC1Command::Execute(llvm::ArrayRef<std::__1::optional<llvm::StringRef>>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>*, bool*) const::$_0::operator()(this=0x000000016fdf96b8) const at Job.cpp:436:34
    frame #28: 0x0000000104bb1e34 clang-dxc`void llvm::function_ref<void ()>::callback_fn<clang::driver::CC1Command::Execute(llvm::ArrayRef<std::__1::optional<llvm::StringRef>>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>*, bool*) const::$_0>(callable=6171891384) at STLFunctionalExtras.h:46:12
    frame #29: 0x00000001011f7984 clang-dxc`llvm::function_ref<void ()>::operator()(this=0x000000016fdf9658) const at STLFunctionalExtras.h:69:12
    frame #30: 0x0000000103184630 clang-dxc`llvm::CrashRecoveryContext::RunSafely(this=0x000000016fdf96f8, Fn=function_ref<void ()> @ 0x000000016fdf9658) at CrashRecoveryContext.cpp:426:3
    frame #31: 0x0000000104bae37c clang-dxc`clang::driver::CC1Command::Execute(this=0x0000000126f24000, Redirects=ArrayRef<std::__1::optional<llvm::StringRef> > @ 0x000000016fdf9740, ErrMsg="", ExecutionFailed=0x000000016fdf9c4f) const at Job.cpp:436:12
    frame #32: 0x0000000104b41b58 clang-dxc`clang::driver::Compilation::ExecuteCommand(this=0x0000000126f22930, C=0x0000000126f24000, FailingCommand=0x000000016fdf9d68, LogOnly=false) const at Compilation.cpp:196:15
    frame #33: 0x0000000104b41de4 clang-dxc`clang::driver::Compilation::ExecuteJobs(this=0x0000000126f22930, Jobs=0x0000000126f229b0, FailingCommands=0x000000016fdfa5f0, LogOnly=false) const at Compilation.cpp:251:19
    frame #34: 0x0000000104b60084 clang-dxc`clang::driver::Driver::ExecuteCompilation(this=0x000000016fdfa640, C=0x0000000126f22930, FailingCommands=0x000000016fdfa5f0) at Driver.cpp:2239:5
    frame #35: 0x0000000100004794 clang-dxc`clang_main(Argc=19, Argv=0x000000016fdfee38, ToolContext=0x000000016fdfe6f8) at driver.cpp:406:21
    frame #36: 0x000000010003f5fc clang-dxc`main(argc=19, argv=0x000000016fdfee38) at clang-driver.cpp:17:10
    frame #37: 0x000000018c812b98 dyld`start + 6076

Module Dump

(lldb) expr M->dump()
; ModuleID = '../DirectML/Product/Shaders/Generated/Quantize_256_8_float16_native_accum32_int8_packed32.hlsl'
source_filename = "../DirectML/Product/Shaders/Generated/Quantize_256_8_float16_native_accum32_int8_packed32.hlsl"
target datalayout = "e-m:e-p:32:32-i1:32-i8:8-i16:16-i32:32-i64:64-f16:16-f32:32-f64:64-n8:16:32:64"
target triple = "dxil-ms-dx"

%"class.hlsl::RWStructuredBuffer.0" = type { target("dx.RawBuffer", i32, 1, 0) }
%"RWStructuredBuffer<half>" = type { half }
%"RWStructuredBuffer<int32_t>" = type { i32 }
%CBuffer.Constants = type { [2 x <4 x i32>], [2 x <4 x i32>], [2 x <4 x i32>], [2 x <4 x i32>], [2 x <4 x i32>], [2 x <4 x i32>], i32, i32, i32, i32 }
%dx.types.Handle = type { ptr }
%dx.types.CBufRet.i32 = type { i32, i32, i32, i32 }
%dx.types.ResRet.f16 = type { half, half, half, half, i32 }
%dx.types.ResRet.i32 = type { i32, i32, i32, i32, i32 }

@_ZL6output = internal unnamed_addr global %"class.hlsl::RWStructuredBuffer.0" poison, align 4
@input = external constant %"RWStructuredBuffer<half>"
@scale = external constant %"RWStructuredBuffer<half>"
@output = external constant %"RWStructuredBuffer<int32_t>"
@Constants = external constant %CBuffer.Constants

; Function Attrs: noinline nounwind memory(readwrite, inaccessiblemem: none)
define void @CSMain() local_unnamed_addr #0 {
entry:
  %agg.tmp22.i63.sroa.0 = alloca target("dx.RawBuffer", i32, 1, 0), align 8
  %0 = call %dx.types.Handle @dx.op.createHandle(i32 57, i8 1, i32 0, i32 0, i1 false) #1
  %1 = call %dx.types.Handle @dx.op.createHandle(i32 57, i8 1, i32 1, i32 1, i1 false) #1
  %2 = call %dx.types.Handle @dx.op.createHandle(i32 57, i8 1, i32 2, i32 2, i1 false) #1
  %3 = call %dx.types.Handle @dx.op.createHandle(i32 57, i8 2, i32 0, i32 0, i1 false) #1
  %4 = call i32 @dx.op.threadId.i32(i32 93, i32 0) #2
  %.load110130 = call %dx.types.CBufRet.i32 @dx.op.cbufferLoadLegacy.i32(i32 59, %dx.types.Handle %3, i32 12) #1
  %.extract111 = extractvalue %dx.types.CBufRet.i32 %.load110130, 3
  %add.i = add i32 %.extract111, %4
  %.load132 = call %dx.types.CBufRet.i32 @dx.op.cbufferLoadLegacy.i32(i32 59, %dx.types.Handle %3, i32 12) #1
  %.extract = extractvalue %dx.types.CBufRet.i32 %.load132, 0
  %cmp.i = icmp ult i32 %add.i, %.extract
  br i1 %cmp.i, label %if.then.i, label %_Z6CSMainDv3_j.exit

if.then.i:                                        ; preds = %entry
  %rem.i = and i32 %add.i, 3
  %.load107131 = call %dx.types.CBufRet.i32 @dx.op.cbufferLoadLegacy.i32(i32 59, %dx.types.Handle %3, i32 12) #1
  %.extract108 = extractvalue %dx.types.CBufRet.i32 %.load107131, 2
  %cmp2.not.i = icmp eq i32 %rem.i, %.extract108
  br i1 %cmp2.not.i, label %if.end.i, label %_Z6CSMainDv3_j.exit

if.end.i:                                         ; preds = %if.then.i
  %agg.tmp.i.sroa.4.0.copyload.load137 = call %dx.types.CBufRet.i32 @dx.op.cbufferLoadLegacy.i32(i32 59, %dx.types.Handle %3, i32 1) #1
  %agg.tmp.i.sroa.4.0.copyload.extract = extractvalue %dx.types.CBufRet.i32 %agg.tmp.i.sroa.4.0.copyload.load137, 0
  %agg.tmp.i.sroa.4.0.copyload.extract86 = extractvalue %dx.types.CBufRet.i32 %agg.tmp.i.sroa.4.0.copyload.load137, 1
  %agg.tmp.i.sroa.4.0.copyload.extract87 = extractvalue %dx.types.CBufRet.i32 %agg.tmp.i.sroa.4.0.copyload.load137, 2
  %agg.tmp.i.sroa.4.0.copyload.extract88 = extractvalue %dx.types.CBufRet.i32 %agg.tmp.i.sroa.4.0.copyload.load137, 3
  %mul.i67 = mul i32 %agg.tmp.i.sroa.4.0.copyload.extract87, %agg.tmp.i.sroa.4.0.copyload.extract88
  %mul14.i = mul i32 %mul.i67, %agg.tmp.i.sroa.4.0.copyload.extract86
  %div6.i.i0 = udiv i32 %add.i, %mul14.i
  %div6.i.i1 = udiv i32 %add.i, %mul.i67
  %div6.i.i2 = udiv i32 %add.i, %agg.tmp.i.sroa.4.0.copyload.extract88
  %div6.i.i3 = udiv i32 %add.i, 1
  %rem8.i.i0 = urem i32 %div6.i.i0, %agg.tmp.i.sroa.4.0.copyload.extract
  %rem8.i.i1 = urem i32 %div6.i.i1, %agg.tmp.i.sroa.4.0.copyload.extract86
  %rem8.i.i2 = urem i32 %div6.i.i2, %agg.tmp.i.sroa.4.0.copyload.extract87
  %rem8.i.i3 = urem i32 %div6.i.i3, %agg.tmp.i.sroa.4.0.copyload.extract88
  %mul21.i = mul i32 %mul14.i, %agg.tmp.i.sroa.4.0.copyload.extract
  %agg.tmp.i.sroa.0.0.copyload.load138 = call %dx.types.CBufRet.i32 @dx.op.cbufferLoadLegacy.i32(i32 59, %dx.types.Handle %3, i32 0) #1
  %agg.tmp.i.sroa.0.0.copyload.extract = extractvalue %dx.types.CBufRet.i32 %agg.tmp.i.sroa.0.0.copyload.load138, 0
  %agg.tmp.i.sroa.0.0.copyload.extract82 = extractvalue %dx.types.CBufRet.i32 %agg.tmp.i.sroa.0.0.copyload.load138, 1
  %agg.tmp.i.sroa.0.0.copyload.extract83 = extractvalue %dx.types.CBufRet.i32 %agg.tmp.i.sroa.0.0.copyload.load138, 2
  %agg.tmp.i.sroa.0.0.copyload.extract84 = extractvalue %dx.types.CBufRet.i32 %agg.tmp.i.sroa.0.0.copyload.load138, 3
  %mul28.i = mul i32 %mul21.i, %agg.tmp.i.sroa.0.0.copyload.extract84
  %mul35.i = mul i32 %mul28.i, %agg.tmp.i.sroa.0.0.copyload.extract83
  %mul42.i = mul i32 %mul35.i, %agg.tmp.i.sroa.0.0.copyload.extract82
  %div.i.i0 = udiv i32 %add.i, %mul42.i
  %div.i.i1 = udiv i32 %add.i, %mul35.i
  %div.i.i2 = udiv i32 %add.i, %mul28.i
  %div.i.i3 = udiv i32 %add.i, %mul21.i
  %rem.i25.i0 = urem i32 %div.i.i0, %agg.tmp.i.sroa.0.0.copyload.extract
  %rem.i25.i1 = urem i32 %div.i.i1, %agg.tmp.i.sroa.0.0.copyload.extract82
  %rem.i25.i2 = urem i32 %div.i.i2, %agg.tmp.i.sroa.0.0.copyload.extract83
  %rem.i25.i3 = urem i32 %div.i.i3, %agg.tmp.i.sroa.0.0.copyload.extract84
  %agg.tmp4.i.sroa.0.0.copyload.load136 = call %dx.types.CBufRet.i32 @dx.op.cbufferLoadLegacy.i32(i32 59, %dx.types.Handle %3, i32 2) #1
  %agg.tmp4.i.sroa.0.0.copyload.extract = extractvalue %dx.types.CBufRet.i32 %agg.tmp4.i.sroa.0.0.copyload.load136, 0
  %agg.tmp4.i.sroa.0.0.copyload.extract90 = extractvalue %dx.types.CBufRet.i32 %agg.tmp4.i.sroa.0.0.copyload.load136, 1
  %agg.tmp4.i.sroa.0.0.copyload.extract91 = extractvalue %dx.types.CBufRet.i32 %agg.tmp4.i.sroa.0.0.copyload.load136, 2
  %agg.tmp4.i.sroa.0.0.copyload.extract92 = extractvalue %dx.types.CBufRet.i32 %agg.tmp4.i.sroa.0.0.copyload.load136, 3
  %agg.tmp4.i.sroa.4.0.copyload.load135 = call %dx.types.CBufRet.i32 @dx.op.cbufferLoadLegacy.i32(i32 59, %dx.types.Handle %3, i32 3) #1
  %agg.tmp4.i.sroa.4.0.copyload.extract = extractvalue %dx.types.CBufRet.i32 %agg.tmp4.i.sroa.4.0.copyload.load135, 0
  %agg.tmp4.i.sroa.4.0.copyload.extract94 = extractvalue %dx.types.CBufRet.i32 %agg.tmp4.i.sroa.4.0.copyload.load135, 1
  %agg.tmp4.i.sroa.4.0.copyload.extract95 = extractvalue %dx.types.CBufRet.i32 %agg.tmp4.i.sroa.4.0.copyload.load135, 2
  %agg.tmp4.i.sroa.4.0.copyload.extract96 = extractvalue %dx.types.CBufRet.i32 %agg.tmp4.i.sroa.4.0.copyload.load135, 3
  %div.i41.i0 = udiv i32 %agg.tmp.i.sroa.0.0.copyload.extract, %agg.tmp4.i.sroa.0.0.copyload.extract
  %div.i41.i1 = udiv i32 %agg.tmp.i.sroa.0.0.copyload.extract82, %agg.tmp4.i.sroa.0.0.copyload.extract90
  %div.i41.i2 = udiv i32 %agg.tmp.i.sroa.0.0.copyload.extract83, %agg.tmp4.i.sroa.0.0.copyload.extract91
  %div.i41.i3 = udiv i32 %agg.tmp.i.sroa.0.0.copyload.extract84, %agg.tmp4.i.sroa.0.0.copyload.extract92
  %div5.i44.i0 = udiv i32 %agg.tmp.i.sroa.4.0.copyload.extract, %agg.tmp4.i.sroa.4.0.copyload.extract
  %div5.i44.i1 = udiv i32 %agg.tmp.i.sroa.4.0.copyload.extract86, %agg.tmp4.i.sroa.4.0.copyload.extract94
  %div5.i44.i2 = udiv i32 %agg.tmp.i.sroa.4.0.copyload.extract87, %agg.tmp4.i.sroa.4.0.copyload.extract95
  %div5.i44.i3 = udiv i32 %agg.tmp.i.sroa.4.0.copyload.extract88, %agg.tmp4.i.sroa.4.0.copyload.extract96
  %div.i32.i0 = udiv i32 %rem.i25.i0, %div.i41.i0
  %div.i32.i1 = udiv i32 %rem.i25.i1, %div.i41.i1
  %div.i32.i2 = udiv i32 %rem.i25.i2, %div.i41.i2
  %div.i32.i3 = udiv i32 %rem.i25.i3, %div.i41.i3
  %div5.i.i0 = udiv i32 %rem8.i.i0, %div5.i44.i0
  %div5.i.i1 = udiv i32 %rem8.i.i1, %div5.i44.i1
  %div5.i.i2 = udiv i32 %rem8.i.i2, %div5.i44.i2
  %div5.i.i3 = udiv i32 %rem8.i.i3, %div5.i44.i3
  %agg.tmp11.i.sroa.0.0.copyload.load134 = call %dx.types.CBufRet.i32 @dx.op.cbufferLoadLegacy.i32(i32 59, %dx.types.Handle %3, i32 6) #1
  %agg.tmp11.i.sroa.0.0.copyload.extract = extractvalue %dx.types.CBufRet.i32 %agg.tmp11.i.sroa.0.0.copyload.load134, 0
  %agg.tmp11.i.sroa.0.0.copyload.extract98 = extractvalue %dx.types.CBufRet.i32 %agg.tmp11.i.sroa.0.0.copyload.load134, 1
  %agg.tmp11.i.sroa.0.0.copyload.extract99 = extractvalue %dx.types.CBufRet.i32 %agg.tmp11.i.sroa.0.0.copyload.load134, 2
  %agg.tmp11.i.sroa.0.0.copyload.extract100 = extractvalue %dx.types.CBufRet.i32 %agg.tmp11.i.sroa.0.0.copyload.load134, 3
  %agg.tmp11.i.sroa.4.0.copyload.load133 = call %dx.types.CBufRet.i32 @dx.op.cbufferLoadLegacy.i32(i32 59, %dx.types.Handle %3, i32 7) #1
  %agg.tmp11.i.sroa.4.0.copyload.extract = extractvalue %dx.types.CBufRet.i32 %agg.tmp11.i.sroa.4.0.copyload.load133, 0
  %agg.tmp11.i.sroa.4.0.copyload.extract102 = extractvalue %dx.types.CBufRet.i32 %agg.tmp11.i.sroa.4.0.copyload.load133, 1
  %agg.tmp11.i.sroa.4.0.copyload.extract103 = extractvalue %dx.types.CBufRet.i32 %agg.tmp11.i.sroa.4.0.copyload.load133, 2
  %agg.tmp11.i.sroa.4.0.copyload.extract104 = extractvalue %dx.types.CBufRet.i32 %agg.tmp11.i.sroa.4.0.copyload.load133, 3
  %5 = mul i32 %div.i32.i0, %agg.tmp11.i.sroa.0.0.copyload.extract
  %dx.mad79124 = call i32 @dx.op.tertiary.i32(i32 49, i32 %div.i32.i1, i32 %agg.tmp11.i.sroa.0.0.copyload.extract98, i32 %5) #2
  %dx.mad80123 = call i32 @dx.op.tertiary.i32(i32 49, i32 %div.i32.i2, i32 %agg.tmp11.i.sroa.0.0.copyload.extract99, i32 %dx.mad79124) #2
  %dx.mad81122 = call i32 @dx.op.tertiary.i32(i32 49, i32 %div.i32.i3, i32 %agg.tmp11.i.sroa.0.0.copyload.extract100, i32 %dx.mad80123) #2
  %6 = mul i32 %div5.i.i0, %agg.tmp11.i.sroa.4.0.copyload.extract
  %dx.mad127 = call i32 @dx.op.tertiary.i32(i32 49, i32 %div5.i.i1, i32 %agg.tmp11.i.sroa.4.0.copyload.extract102, i32 %6) #2
  %dx.mad77126 = call i32 @dx.op.tertiary.i32(i32 49, i32 %div5.i.i2, i32 %agg.tmp11.i.sroa.4.0.copyload.extract103, i32 %dx.mad127) #2
  %dx.mad78125 = call i32 @dx.op.tertiary.i32(i32 49, i32 %div5.i.i3, i32 %agg.tmp11.i.sroa.4.0.copyload.extract104, i32 %dx.mad77126) #2
  %add.i62 = add i32 %dx.mad78125, %dx.mad81122
  %7 = call %dx.types.ResRet.f16 @dx.op.rawBufferLoad.f16(i32 139, %dx.types.Handle %0, i32 %add.i, i32 0, i8 1, i32 2)
  %8 = extractvalue %dx.types.ResRet.f16 %7, 0
  %9 = call %dx.types.ResRet.f16 @dx.op.rawBufferLoad.f16(i32 139, %dx.types.Handle %1, i32 %add.i62, i32 0, i8 1, i32 2)
  %10 = extractvalue %dx.types.ResRet.f16 %9, 0
  %conv.i = fpext reassoc nnan ninf nsz arcp afn half %8 to float
  %conv18.i = fpext reassoc nnan ninf nsz arcp afn half %10 to float
  %div.i.i = fdiv reassoc nnan ninf nsz arcp afn float %conv.i, %conv18.i
  %elt.roundeven.i.i121 = call float @dx.op.unary.f32(i32 26, float %div.i.i) #2
  %dx.max128 = call float @dx.op.binary.f32(i32 35, float %elt.roundeven.i.i121, float -1.280000e+02) #2
  %dx.min129 = call float @dx.op.binary.f32(i32 36, float %dx.max128, float 1.270000e+02) #2
  %conv21.i = fptosi float %dx.min129 to i32
  %11 = bitcast ptr @_ZL6output to ptr
  %12 = load i32, ptr %11, align 4
  %13 = bitcast ptr %agg.tmp22.i63.sroa.0 to ptr
  store target("dx.RawBuffer", i32, 1, 0) undef, ptr %13, align 4
  %14 = bitcast ptr %agg.tmp22.i63.sroa.0 to ptr
  store i32 %12, ptr %14, align 4
  %div9.i = lshr i32 %add.i, 2
  %rem.i64 = shl i32 %add.i, 3
  %mul.i = and i32 %rem.i64, 24
  %shl.i = shl nuw i32 255, %mul.i
  %15 = call %dx.types.ResRet.i32 @dx.op.rawBufferLoad.i32(i32 139, %dx.types.Handle %2, i32 %div9.i, i32 0, i8 1, i32 4)
  %16 = extractvalue %dx.types.ResRet.i32 %15, 0
  %shl2.i = shl i32 %conv21.i, %mul.i
  %xor.i = xor i32 %16, %shl2.i
  %and.i = and i32 %xor.i, %shl.i
  %xor3.i = xor i32 %and.i, %16
  call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %div9.i, i32 0, i32 %xor3.i, i32 undef, i32 undef, i32 undef, i8 1, i32 4)
  %17 = bitcast ptr %agg.tmp22.i63.sroa.0 to ptr
  store target("dx.RawBuffer", i32, 1, 0) undef, ptr %17, align 4
  br label %_Z6CSMainDv3_j.exit

_Z6CSMainDv3_j.exit:                              ; preds = %if.then.i, %entry, %if.end.i
  ret void
}

declare float @dx.op.unary.f32(i32 %0, float %1)

declare i32 @dx.op.threadId.i32(i32 %0, i32 %1)

declare %dx.types.Handle @dx.op.createHandle(i32 %0, i8 %1, i32 %2, i32 %3, i1 %4)

declare %dx.types.ResRet.f16 @dx.op.rawBufferLoad.f16(i32 %0, %dx.types.Handle %1, i32 %2, i32 %3, i8 %4, i32 %5)

declare %dx.types.ResRet.i32 @dx.op.rawBufferLoad.i32(i32 %0, %dx.types.Handle %1, i32 %2, i32 %3, i8 %4, i32 %5)

declare void @dx.op.rawBufferStore.i32(i32 %0, %dx.types.Handle %1, i32 %2, i32 %3, i32 %4, i32 %5, i32 %6, i32 %7, i8 %8, i32 %9)

declare i32 @dx.op.tertiary.i32(i32 %0, i32 %1, i32 %2, i32 %3)

declare float @dx.op.binary.f32(i32 %0, float %1, float %2)

declare %dx.types.CBufRet.i32 @dx.op.cbufferLoadLegacy.i32(i32 %0, %dx.types.Handle %1, i32 %2)

attributes #0 = { noinline nounwind memory(readwrite, inaccessiblemem: none) }
attributes #1 = { memory(read) }
attributes #2 = { memory(none) }

!dx.valver = !{!0}
!llvm.ident = !{!1}
!dx.shaderModel = !{!2}
!dx.version = !{!3}
!dx.resources = !{!4}
!dx.entryPoints = !{!12}
!llvm.module.flags = !{!15, !16, !17}

!0 = !{i32 1, i32 8}
!1 = !{!"clang version 22.0.0git (git@github.com:llvm/llvm-project.git 1e78354d88979fd524f2b4626249c978b2d5ebaa)"}
!2 = !{!"cs", i32 6, i32 2}
!3 = !{i32 1, i32 2}
!4 = !{null, !5, !10, null}
!5 = !{!6, !7, !8}
!6 = !{i32 0, ptr @input, !"input", i32 0, i32 0, i32 1, i32 12, i1 false, i1 false, i1 false, !3}
!7 = !{i32 1, ptr @scale, !"scale", i32 0, i32 1, i32 1, i32 12, i1 false, i1 false, i1 false, !3}
!8 = !{i32 2, ptr @output, !"output", i32 0, i32 2, i32 1, i32 12, i1 false, i1 false, i1 false, !9}
!9 = !{i32 1, i32 4}
!10 = !{!11}
!11 = !{i32 0, ptr @Constants, !"Constants", i32 0, i32 0, i32 1, i32 208, null}
!12 = !{ptr @CSMain, !"CSMain", null, !4, !13}
!13 = !{i32 0, i64 8388656, i32 4, !14}
!14 = !{i32 256, i32 1, i32 1}
!15 = !{i32 1, !"wchar_size", i32 4}
!16 = !{i32 1, !"dx.nativelowprec", i32 1}
!17 = !{i32 2, !"frame-pointer", i32 2}
  Evaluated this expression after applying Fix-It(s):
    M.dump()

[farzonl]

Assuming code is the same:

DXC generates IR that looks like this

define void @CSMain() {
  %dest_UAV_structbuf = call %dx.types.Handle @dx.op.createHandle(i32 57, i8 1, i32 0, i32 3, i1 false)
  %1 = call i32 @dx.op.threadId.i32(i32 93, i32 0)
  call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %dest_UAV_structbuf, i32 %1, i32 0, i32 0, i32 undef, i32 undef, i32 undef, i8 1, i32 4)
  ret void
}

And clang generates code that looks like this:

define void @CSMain() {
entry:
  %agg.tmp.i1.sroa.0 = alloca target("dx.RawBuffer", i32, 1, 0), align 8
  %0 = call %dx.types.Handle @dx.op.createHandle(i32 57, i8 1, i32 0, i32 3, i1 false) #1
  %1 = call i32 @dx.op.threadId.i32(i32 93, i32 0) #2
  %2 = bitcast ptr @_ZL4dest to ptr
  %3 = load i32, ptr %2, align 4
  %4 = bitcast ptr %agg.tmp.i1.sroa.0 to ptr
  store target("dx.RawBuffer", i32, 1, 0) undef, ptr %4, align 4
  %5 = bitcast ptr %agg.tmp.i1.sroa.0 to ptr
  store i32 %3, ptr %5, align 4
  call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %0, i32 %1, i32 0, i32 0, i32 undef, i32 undef, i32 undef, i8 1, i32 4)
  %6 = bitcast ptr %agg.tmp.i1.sroa.0 to ptrstore target("dx.RawBuffer", i32, 1, 0) undef, ptr %6, align 4
  ret void
}

That makes me think we are going to need our own custom optimization passes to pull off similar code gen?

One fix seems to be to optimize.

[damyanp]

If the resolution to this involves significant work then we may want to track that with separate tasks. Seems there's an open question at the moment about whether this is a more simple bug in dxil op lowering or a resource pass rather than a missing optimization.

[farzonl]

This is fixed if we run Dead Store Elimination (DSE) after the forward handle pass.

../llvm_debug_build/bin/opt -mtriple=dxil--shadermodel6.3-compute -S -passes='function(dxil-forward-handle-accesses),dse' llvm/test/CodeGen/DirectX/issue-140819_allow_forward_handle_on_alloca.ll

code gen after

define void @CSMain() local_unnamed_addr {
entry:
  %handle = tail call target("dx.RawBuffer", i32, 1, 0) @llvm.dx.resource.handlefrombinding.tdx.RawBuffer_i32_1_0t(i32 0, i32 3, i32 1, i32 0, i1 false, ptr nonnull @name)
  store target("dx.RawBuffer", i32, 1, 0) %handle, ptr @global, align 4
  %buff = tail call noundef nonnull align 4 dereferenceable(4) ptr @llvm.dx.resource.getpointer.p0.tdx.RawBuffer_i32_1_0t(target("dx.RawBuffer", i32, 1, 0) %handle, i32 0)
  store i32 0, ptr %buff, align 4
  ret void
}

The question now is do we want to bring back the legacy DSE pass apis.

[bogner]

Yes, I think running DSE makes sense here. Tons of passes still have the legacy pass APIs (they just removed all of the ones that didn't happen to be used in tree at some point...) so I don't think there's a real cost to bringing it back.

It would, of course, be nice to be able to use the new pass manager for a backend's passes, but the infrastructure for that isn't there yet.