From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001 From: David Neto Date: Tue, 16 Jul 2024 16:34:52 -0400 Subject: Add CMake option DXC_CODEGEN_EXCEPTIONS_TRAP (#6764) When enabled, any hlsl::Exception thrown during code generation and optimization will cause the process to trap. Bug: 346618785 Change-Id: I9ac966d339ec3090e3455d51a9d7516cc5a3f153 HLMatrixLower: allow exceptions to propagate out (#6710) If an exception is thrown, don't block it in the TempOverloadPool destructor. Allow it to propagate out as a user-visible error. Explicitly clear the TempOverloadPool before returning from the HLMatrixLowerPass::runOnModule. In the normal case, when no exception is thrown, that will still verify that all the overloads actually have been lowered, and will assert out if they aren't. Bug: 346618785 Change-Id: Id421ca324e4d799477a41abb14b966e8f39be482 Reviewed-on: https://chromium-review.googlesource.com/c/external/github.com/microsoft/DirectXShaderCompiler/+/5715228 Reviewed-by: Natalie Chouinard diff --git a/CMakeLists.txt b/CMakeLists.txt index 15097b8473bf9f6c2b6c28ac095936616aa917f6..ac35ebd1a6afd55934da502ddb0e0c286da10db8 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -104,6 +104,9 @@ endif() option(DXC_DISABLE_ALLOCATOR_OVERRIDES "Disable usage of allocator overrides" OFF) mark_as_advanced(DXC_DISABLE_ALLOCATOR_OVERRIDES) +option(DXC_CODEGEN_EXCEPTIONS_TRAP "An exception in code generation generates a trap, ending the compiler process" OFF) +mark_as_advanced(DXC_CODEGEN_EXCEPTIONS_TRAP) + # adjust link option to enable debugging from kernel mode; not compatible with incremental linking if(NOT CMAKE_VERSION VERSION_LESS "3.13" AND WIN32 AND NOT CMAKE_C_COMPILER_ARCHITECTURE_ID STREQUAL "ARM64EC") add_link_options(/DEBUGTYPE:CV,FIXUP,PDATA /INCREMENTAL:NO) diff --git a/include/dxc/config.h.cmake b/include/dxc/config.h.cmake index 6d7450323ec864ef5f6e8796c46423d49f278b76..7226ed75a108f63c3c894b59eb7e0dbc5fb66002 100644 --- a/include/dxc/config.h.cmake +++ b/include/dxc/config.h.cmake @@ -1,2 +1,4 @@ /* Disable overriding memory allocators. */ #cmakedefine DXC_DISABLE_ALLOCATOR_OVERRIDES +/* Generate a trap if an hlsl::Exception is thrown during code generation */ +#cmakedefine DXC_CODEGEN_EXCEPTIONS_TRAP diff --git a/lib/HLSL/HLMatrixLowerPass.cpp b/lib/HLSL/HLMatrixLowerPass.cpp index d5959eb9335465f67d8e7ef7d7ab4eb720274226..3bb9f65834ecfd0c7a25c8a176f62c124e9967e5 100644 --- a/lib/HLSL/HLMatrixLowerPass.cpp +++ b/lib/HLSL/HLMatrixLowerPass.cpp @@ -60,7 +60,12 @@ class TempOverloadPool { public: TempOverloadPool(llvm::Module &Module, const char *BaseName) : Module(Module), BaseName(BaseName) {} - ~TempOverloadPool() { clear(); } + ~TempOverloadPool() { + if (!Funcs.empty()) { + // The flow has thrown an exception. Let that exception + // propagate out and be reported as a compile error. + } + } Function *get(FunctionType *Ty); bool contains(FunctionType *Ty) const { return Funcs.count(Ty) != 0; } @@ -248,12 +253,15 @@ bool HLMatrixLowerPass::runOnModule(Module &M) { m_matToVecStubs = nullptr; m_vecToMatStubs = nullptr; - // If you hit an assert during TempOverloadPool destruction, + // If you hit an assert while clearing TempOverloadPool, // it means that either a matrix producer was lowered, // causing a translation stub to be created, // but the consumer of that matrix was never (properly) lowered. // Or the opposite: a matrix consumer was lowered and not its producer. + matToVecStubs.clear(); + vecToMatStubs.clear(); + return true; } diff --git a/tools/clang/lib/CodeGen/BackendUtil.cpp b/tools/clang/lib/CodeGen/BackendUtil.cpp index 294ca05946eed6d87a3ad4e2b56a641d24065760..1e9adb40f4bae4cada42b5581582d50dc3676594 100644 --- a/tools/clang/lib/CodeGen/BackendUtil.cpp +++ b/tools/clang/lib/CodeGen/BackendUtil.cpp @@ -8,6 +8,10 @@ //===----------------------------------------------------------------------===// #include "clang/CodeGen/BackendUtil.h" +#include "dxc/HLSL/DxilGenerationPass.h" // HLSL Change +#include "dxc/HLSL/HLMatrixLowerPass.h" // HLSL Change +#include "dxc/Support/Global.h" // HLSL Change +#include "dxc/config.h" // HLSL Change #include "clang/Basic/Diagnostic.h" #include "clang/Basic/LangOptions.h" #include "clang/Basic/TargetOptions.h" @@ -41,10 +45,8 @@ #include "llvm/Transforms/ObjCARC.h" #include "llvm/Transforms/Scalar.h" #include "llvm/Transforms/Utils/SymbolRewriter.h" +#include #include -#include "dxc/HLSL/DxilGenerationPass.h" // HLSL Change -#include "dxc/HLSL/HLMatrixLowerPass.h" // HLSL Change -#include "dxc/Support/Global.h" // HLSL Change using namespace clang; using namespace llvm; @@ -780,6 +782,13 @@ void clang::EmitBackendOutput(DiagnosticsEngine &Diags, } catch (const ::hlsl::Exception &hlslException) { Diags.Report(Diags.getCustomDiagID(DiagnosticsEngine::Error, "%0\n")) << StringRef(hlslException.what()); +#if defined(DXC_CODEGEN_EXCEPTIONS_TRAP) + // llvm::errs() doesn't work in release builds on Linux. + // Use C-style fprintf because it works everywhere. + fprintf(stderr, "internal codegen error: %s\n", hlslException.what()); + fflush(stderr); + LLVM_BUILTIN_TRAP; +#endif } // HLSL Change Ends // If an optional clang TargetInfo description string was passed in, use it to diff --git a/tools/clang/test/DXC/Passes/HLMatrixLower/dont_crash_on_invalid_cast.ll b/tools/clang/test/DXC/Passes/HLMatrixLower/dont_crash_on_invalid_cast.ll new file mode 100644 index 0000000000000000000000000000000000000000..fca52d11595d3ac99dedb28219f0746a26add6c9 --- /dev/null +++ b/tools/clang/test/DXC/Passes/HLMatrixLower/dont_crash_on_invalid_cast.ll @@ -0,0 +1,130 @@ +; RUN: not %dxopt %s -hlsl-passes-resume -hlmatrixlower -S | FileCheck %s + + +; The HL matrix lowering pass can sometimes throw an exception +; due to an invalid LLVM-level cast call. Make sure that +; propagates out to a user-level error. + +; Note: There is still a bug in the compiler here. Not all matrix +; lowerings are covered by the pass. + +; TODO: Fix the underlying bug https://github.com/microsoft/DirectXShaderCompiler/issues/6723 +; Once that is fixed, this test changes or should be deleted. + + +; CHECK: Operation failed - error code + +; +; Buffer Definitions: +; +; cbuffer $Globals +; { +; +; [0 x i8] (type annotation not present) +; +; } +; +; +; Resource Bindings: +; +; Name Type Format Dim ID HLSL Bind Count +; ------------------------------ ---------- ------- ----------- ------- -------------- ------ +; $Globals cbuffer NA NA CB0 cb4294967295 1 +; +target datalayout = "e-m:e-p:32:32-i1:32-i8:32-i16:32-i32:32-i64:64-f16:32-f32:32-f64:64-n8:16:32:64" +target triple = "dxil-ms-dx" + +%ConstantBuffer = type opaque +%class.matrix.float.2.4 = type { [2 x <4 x float>] } +%struct.e = type { [1 x %struct.d], [1 x %struct.d] } +%struct.d = type { %struct.a } +%struct.a = type { float, %class.matrix.float.2.4 } + +@"$Globals" = external constant %ConstantBuffer +@g.0.0.0 = internal global [1 x float] undef, align 4 +@g.0.0.1 = internal global [1 x %class.matrix.float.2.4] undef, align 4 +@g.1.0.0 = internal global [1 x float] undef, align 4 +@g.1.0.1 = internal global [1 x %class.matrix.float.2.4] undef, align 4 + +; Function Attrs: nounwind +define void @main() #0 { +entry: + %h.0.1 = alloca %class.matrix.float.2.4, !dbg !26 ; line:13 col:17 + store float 0.000000e+00, float* getelementptr inbounds ([1 x float], [1 x float]* @g.0.0.0, i32 0, i32 0), !dbg !26 ; line:13 col:17 + %0 = call %class.matrix.float.2.4 @"dx.hl.init.rn.%class.matrix.float.2.4 (i32, <8 x float>)"(i32 0, <8 x float> zeroinitializer) #0, !dbg !26 ; line:13 col:17 + %1 = call %class.matrix.float.2.4 @"dx.hl.cast.rowMatToColMat.%class.matrix.float.2.4 (i32, %class.matrix.float.2.4)"(i32 7, %class.matrix.float.2.4 %0) #0, !dbg !26 ; line:13 col:17 + %2 = call %class.matrix.float.2.4 @"dx.hl.matldst.colStore.%class.matrix.float.2.4 (i32, %class.matrix.float.2.4*, %class.matrix.float.2.4)"(i32 1, %class.matrix.float.2.4* getelementptr inbounds ([1 x %class.matrix.float.2.4], [1 x %class.matrix.float.2.4]* @g.0.0.1, i32 0, i32 0), %class.matrix.float.2.4 %1) #0, !dbg !26 ; line:13 col:17 + store float 0.000000e+00, float* getelementptr inbounds ([1 x float], [1 x float]* @g.1.0.0, i32 0, i32 0), !dbg !26 ; line:13 col:17 + %3 = call %class.matrix.float.2.4 @"dx.hl.init.rn.%class.matrix.float.2.4 (i32, <8 x float>)"(i32 0, <8 x float> zeroinitializer) #0, !dbg !26 ; line:13 col:17 + %4 = call %class.matrix.float.2.4 @"dx.hl.cast.rowMatToColMat.%class.matrix.float.2.4 (i32, %class.matrix.float.2.4)"(i32 7, %class.matrix.float.2.4 %3) #0, !dbg !26 ; line:13 col:17 + %5 = call %class.matrix.float.2.4 @"dx.hl.matldst.colStore.%class.matrix.float.2.4 (i32, %class.matrix.float.2.4*, %class.matrix.float.2.4)"(i32 1, %class.matrix.float.2.4* getelementptr inbounds ([1 x %class.matrix.float.2.4], [1 x %class.matrix.float.2.4]* @g.1.0.1, i32 0, i32 0), %class.matrix.float.2.4 %4) #0, !dbg !26 ; line:13 col:17 + %6 = load float, float* getelementptr inbounds ([1 x float], [1 x float]* @g.1.0.0, i32 0, i32 0), !dbg !32 ; line:17 col:9 + %7 = getelementptr inbounds %class.matrix.float.2.4, %class.matrix.float.2.4* %h.0.1, i32 0, i32 0, i32 0, !dbg !32 ; line:17 col:9 + %8 = load <4 x float>, <4 x float>* getelementptr inbounds ([1 x %class.matrix.float.2.4], [1 x %class.matrix.float.2.4]* @g.1.0.1, i32 0, i32 0, i32 0, i32 0), !dbg !32 ; line:17 col:9 + store <4 x float> %8, <4 x float>* %7, !dbg !32 ; line:17 col:9 + %9 = getelementptr inbounds %class.matrix.float.2.4, %class.matrix.float.2.4* %h.0.1, i32 0, i32 0, i32 1, !dbg !32 ; line:17 col:9 + %10 = load <4 x float>, <4 x float>* getelementptr inbounds ([1 x %class.matrix.float.2.4], [1 x %class.matrix.float.2.4]* @g.1.0.1, i32 0, i32 0, i32 0, i32 1), !dbg !32 ; line:17 col:9 + store <4 x float> %10, <4 x float>* %9, !dbg !32 ; line:17 col:9 + ret void, !dbg !33 ; line:18 col:3 +} + +; Function Attrs: nounwind +declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture readonly, i64, i32, i1) #0 + +; Function Attrs: nounwind readnone +declare %class.matrix.float.2.4 @"dx.hl.init.rn.%class.matrix.float.2.4 (i32, <8 x float>)"(i32, <8 x float>) #1 + +; Function Attrs: nounwind readnone +declare %class.matrix.float.2.4 @"dx.hl.cast.rowMatToColMat.%class.matrix.float.2.4 (i32, %class.matrix.float.2.4)"(i32, %class.matrix.float.2.4) #1 + +; Function Attrs: nounwind +declare %class.matrix.float.2.4 @"dx.hl.matldst.colStore.%class.matrix.float.2.4 (i32, %class.matrix.float.2.4*, %class.matrix.float.2.4)"(i32, %class.matrix.float.2.4*, %class.matrix.float.2.4) #0 + +attributes #0 = { nounwind } +attributes #1 = { nounwind readnone } + +!llvm.module.flags = !{!0} +!pauseresume = !{!1} +!llvm.ident = !{!2} +!dx.version = !{!3} +!dx.valver = !{!4} +!dx.shaderModel = !{!5} +!dx.typeAnnotations = !{!6, !15} +!dx.entryPoints = !{!19} +!dx.fnprops = !{!23} +!dx.options = !{!24, !25} + +!0 = !{i32 2, !"Debug Info Version", i32 3} +!1 = !{!"hlsl-hlemit", !"hlsl-hlensure"} +!2 = !{!"dxc(private) 1.8.0.4640 (issue-785, 45018c752d)"} +!3 = !{i32 1, i32 0} +!4 = !{i32 1, i32 8} +!5 = !{!"cs", i32 6, i32 0} +!6 = !{i32 0, %struct.e undef, !7, %struct.d undef, !10, %struct.a undef, !11} +!7 = !{i32 152, !8, !9} +!8 = !{i32 6, !"c", i32 3, i32 0} +!9 = !{i32 6, !"f", i32 3, i32 80} +!10 = !{i32 72, !8} +!11 = !{i32 72, !12, !13} +!12 = !{i32 6, !"b", i32 3, i32 0, i32 7, i32 9} +!13 = !{i32 6, !"c", i32 2, !14, i32 3, i32 16, i32 7, i32 9} +!14 = !{i32 2, i32 4, i32 2} +!15 = !{i32 1, void ()* @main, !16} +!16 = !{!17} +!17 = !{i32 1, !18, !18} +!18 = !{} +!19 = !{void ()* @main, !"main", null, !20, null} +!20 = !{null, null, !21, null} +!21 = !{!22} +!22 = !{i32 0, %ConstantBuffer* @"$Globals", !"$Globals", i32 0, i32 -1, i32 1, i32 0, null} +!23 = !{void ()* @main, i32 5, i32 1, i32 1, i32 1} +!24 = !{i32 64} +!25 = !{i32 -1} +!26 = !DILocation(line: 13, column: 17, scope: !27, inlinedAt: !30) +!27 = !DISubprogram(name: "??__Eg@@YAXXZ", scope: !28, file: !28, line: 13, type: !29, isLocal: true, isDefinition: true, scopeLine: 13, flags: DIFlagPrototyped, isOptimized: false) +!28 = !DIFile(filename: "a.hlsl", directory: "") +!29 = !DISubroutineType(types: !18) +!30 = distinct !DILocation(line: 16, scope: !31) +!31 = !DISubprogram(name: "main", scope: !28, file: !28, line: 16, type: !29, isLocal: false, isDefinition: true, scopeLine: 16, flags: DIFlagPrototyped, isOptimized: false, function: void ()* @main) +!32 = !DILocation(line: 17, column: 9, scope: !31) +!33 = !DILocation(line: 18, column: 3, scope: !31)