| 1 | //===----------------------------------------------------------------------===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | |
| 9 | #include "WebAssembly.h" |
| 10 | #include "WebAssemblyTargetMachine.h" |
| 11 | #include "llvm/CodeGen/AtomicExpand.h" |
| 12 | #include "llvm/CodeGen/IndirectBrExpand.h" |
| 13 | #include "llvm/CodeGen/MachineBlockPlacement.h" |
| 14 | #include "llvm/CodeGen/MachineCopyPropagation.h" |
| 15 | #include "llvm/CodeGen/MachineLateInstrsCleanup.h" |
| 16 | #include "llvm/CodeGen/PatchableFunction.h" |
| 17 | #include "llvm/CodeGen/PostRAMachineSink.h" |
| 18 | #include "llvm/CodeGen/PostRASchedulerList.h" |
| 19 | #include "llvm/CodeGen/RegisterCoalescerPass.h" |
| 20 | #include "llvm/CodeGen/RemoveLoadsIntoFakeUses.h" |
| 21 | #include "llvm/CodeGen/ShrinkWrap.h" |
| 22 | #include "llvm/CodeGen/UnreachableBlockElim.h" |
| 23 | #include "llvm/IR/PassInstrumentation.h" |
| 24 | #include "llvm/MC/MCStreamer.h" |
| 25 | #include "llvm/Passes/CodeGenPassBuilder.h" |
| 26 | #include "llvm/Passes/PassBuilder.h" |
| 27 | #include "llvm/Support/CodeGen.h" |
| 28 | #include "llvm/Target/CGPassBuilderOption.h" |
| 29 | #include "llvm/Transforms/Utils/LowerGlobalDtors.h" |
| 30 | #include "llvm/Transforms/Utils/LowerInvoke.h" |
| 31 | |
| 32 | using namespace llvm; |
| 33 | |
| 34 | namespace WebAssembly { |
| 35 | extern cl::opt<bool> WasmDisableExplicitLocals; |
| 36 | extern cl::opt<bool> WasmEnableEH; |
| 37 | extern cl::opt<bool> WasmEnableEmEH; |
| 38 | extern cl::opt<bool> WasmEnableEmSjLj; |
| 39 | extern cl::opt<bool> WasmEnableSjLj; |
| 40 | } // namespace WebAssembly |
| 41 | |
| 42 | using llvm::WebAssembly::WasmDisableExplicitLocals; |
| 43 | using llvm::WebAssembly::WasmEnableEH; |
| 44 | using llvm::WebAssembly::WasmEnableEmEH; |
| 45 | using llvm::WebAssembly::WasmEnableEmSjLj; |
| 46 | using llvm::WebAssembly::WasmEnableSjLj; |
| 47 | |
| 48 | namespace { |
| 49 | |
| 50 | class WebAssemblyCodeGenPassBuilder |
| 51 | : public CodeGenPassBuilder<WebAssemblyCodeGenPassBuilder, |
| 52 | WebAssemblyTargetMachine> { |
| 53 | using Base = CodeGenPassBuilder<WebAssemblyCodeGenPassBuilder, |
| 54 | WebAssemblyTargetMachine>; |
| 55 | |
| 56 | public: |
| 57 | explicit WebAssemblyCodeGenPassBuilder(WebAssemblyTargetMachine &TM, |
| 58 | const CGPassBuilderOption &Opts, |
| 59 | PassInstrumentationCallbacks *PIC) |
| 60 | : CodeGenPassBuilder(TM, Opts, PIC) { |
| 61 | disablePass<MachineLateInstrsCleanupPass, MachineCopyPropagationPass, |
| 62 | PostRAMachineSinkingPass, PostRASchedulerPass, |
| 63 | FuncletLayoutPass, StackMapLivenessPass, PatchableFunctionPass, |
| 64 | ShrinkWrapPass, RemoveLoadsIntoFakeUsesPass, |
| 65 | MachineBlockPlacementPass>(); |
| 66 | |
| 67 | // Currently RegisterCoalesce degrades wasm debug info quality by a |
| 68 | // significant margin. As a quick fix, disable this for -O1, which is often |
| 69 | // used for debugging large applications. Disabling this increases code size |
| 70 | // of Emscripten core benchmarks by ~5%, which is acceptable for -O1, which |
| 71 | // is usually not used for production builds. |
| 72 | // TODO Investigate why RegisterCoalesce degrades debug info quality and fix |
| 73 | // it properly |
| 74 | if (getOptLevel() == CodeGenOptLevel::Less) |
| 75 | disablePass<RegisterCoalescerPass>(); |
| 76 | } |
| 77 | |
| 78 | void addIRPasses(PassManagerWrapper &PMW) const; |
| 79 | void addISelPrepare(PassManagerWrapper &PMW) const; |
| 80 | Error addInstSelector(PassManagerWrapper &PMW) const; |
| 81 | void addPreEmitPass(PassManagerWrapper &PMW) const; |
| 82 | }; |
| 83 | |
| 84 | void WebAssemblyCodeGenPassBuilder::addIRPasses(PassManagerWrapper &PMW) const { |
| 85 | // Add signatures to prototype-less function declarations |
| 86 | flushFPMsToMPM(PMW); |
| 87 | addModulePass(Pass: WebAssemblyAddMissingPrototypesPass(), PMW); |
| 88 | |
| 89 | // Lower .llvm.global_dtors into .llvm.global_ctors with __cxa_atexit calls. |
| 90 | addModulePass(Pass: LowerGlobalDtorsPass(), PMW); |
| 91 | |
| 92 | // Fix function bitcasts, as WebAssembly requires caller and callee signatures |
| 93 | // to match. |
| 94 | addModulePass(Pass: WebAssemblyFixFunctionBitcastsPass(), PMW); |
| 95 | |
| 96 | // Optimize "returned" function attributes. |
| 97 | if (getOptLevel() != CodeGenOptLevel::None) |
| 98 | addFunctionPass(Pass: WebAssemblyOptimizeReturnedPass(), PMW); |
| 99 | |
| 100 | // If exception handling is not enabled and setjmp/longjmp handling is |
| 101 | // enabled, we lower invokes into calls and delete unreachable landingpad |
| 102 | // blocks. Lowering invokes when there is no EH support is done in |
| 103 | // TargetPassConfig::addPassesToHandleExceptions, but that runs after these IR |
| 104 | // passes and Emscripten SjLj handling expects all invokes to be lowered |
| 105 | // before. |
| 106 | if (!WasmEnableEmEH && !WasmEnableEH) { |
| 107 | addFunctionPass(Pass: LowerInvokePass(), PMW); |
| 108 | // The lower invoke pass may create unreachable code. Remove it in order not |
| 109 | // to process dead blocks in setjmp/longjmp handling. |
| 110 | addFunctionPass(Pass: UnreachableBlockElimPass(), PMW); |
| 111 | } |
| 112 | |
| 113 | // Handle exceptions and setjmp/longjmp if enabled. Unlike Wasm EH preparation |
| 114 | // done in WasmEHPrepare pass, Wasm SjLj preparation shares libraries and |
| 115 | // transformation algorithms with Emscripten SjLj, so we run |
| 116 | // LowerEmscriptenEHSjLj pass also when Wasm SjLj is enabled. |
| 117 | if (WasmEnableEmEH || WasmEnableEmSjLj || WasmEnableSjLj) { |
| 118 | flushFPMsToMPM(PMW); |
| 119 | addModulePass(Pass: WebAssemblyLowerEmscriptenEHSjLjPass(), PMW); |
| 120 | } |
| 121 | |
| 122 | // Expand indirectbr instructions to switches. |
| 123 | addFunctionPass(Pass: IndirectBrExpandPass(TM), PMW); |
| 124 | |
| 125 | // Try to expand `vecreduce_{and, or}` into `{any, all}_true`. |
| 126 | addFunctionPass(Pass: WebAssemblyReduceToAnyAllTruePass(TM), PMW); |
| 127 | |
| 128 | Base::addIRPasses(PMW); |
| 129 | } |
| 130 | |
| 131 | void WebAssemblyCodeGenPassBuilder::addISelPrepare( |
| 132 | PassManagerWrapper &PMW) const { |
| 133 | // We need to move reference type allocas to WASM_ADDRESS_SPACE_VAR so that |
| 134 | // loads and stores are promoted to local.gets/local.sets. |
| 135 | addFunctionPass(Pass: WebAssemblyRefTypeMem2LocalPass(), PMW); |
| 136 | // Lower atomics and TLS if necessary |
| 137 | flushFPMsToMPM(PMW); |
| 138 | addModulePass(Pass: WebAssemblyCoalesceFeaturesAndStripAtomicsPass(TM), PMW); |
| 139 | |
| 140 | // This is a no-op if atomics are not used in the module |
| 141 | addFunctionPass(Pass: AtomicExpandPass(TM), PMW); |
| 142 | |
| 143 | Base::addISelPrepare(PMW); |
| 144 | } |
| 145 | |
| 146 | Error WebAssemblyCodeGenPassBuilder::addInstSelector( |
| 147 | PassManagerWrapper &PMW) const { |
| 148 | addMachineFunctionPass(Pass: WebAssemblyISelDAGToDAGPass(TM, getOptLevel()), PMW); |
| 149 | |
| 150 | // Run the argument-move pass immediately after the ScheduleDAG scheduler |
| 151 | // so that we can fix up the ARGUMENT instructions before anything else |
| 152 | // sees them in the wrong place. |
| 153 | addMachineFunctionPass(Pass: WebAssemblyArgumentMovePass(), PMW); |
| 154 | |
| 155 | // Set the p2align operands. This information is present during ISel, however |
| 156 | // it's inconvenient to collect. Collect it now, and update the immediate |
| 157 | // operands. |
| 158 | addMachineFunctionPass(Pass: WebAssemblySetP2AlignOperandsPass(), PMW); |
| 159 | |
| 160 | // Eliminate range checks and add default targets to br_table instructions. |
| 161 | addMachineFunctionPass(Pass: WebAssemblyFixBrTableDefaultsPass(), PMW); |
| 162 | |
| 163 | // unreachable is terminator, non-terminator instruction after it is not |
| 164 | // allowed. |
| 165 | addMachineFunctionPass(Pass: WebAssemblyCleanCodeAfterTrapPass(), PMW); |
| 166 | |
| 167 | return Error::success(); |
| 168 | } |
| 169 | |
| 170 | void WebAssemblyCodeGenPassBuilder::addPreEmitPass( |
| 171 | PassManagerWrapper &PMW) const { |
| 172 | Base::addPreEmitPass(PMW); |
| 173 | |
| 174 | // Nullify DBG_VALUE_LISTs that we cannot handle. |
| 175 | addMachineFunctionPass(Pass: WebAssemblyNullifyDebugValueListsPass(), PMW); |
| 176 | |
| 177 | // Remove any unreachable blocks that may be left floating around. |
| 178 | // Rare, but possible. Needed for WebAssemblyFixIrreducibleControlFlow. |
| 179 | addMachineFunctionPass(Pass: UnreachableMachineBlockElimPass(), PMW); |
| 180 | |
| 181 | // Eliminate multiple-entry loops. |
| 182 | addMachineFunctionPass(Pass: WebAssemblyFixIrreducibleControlFlowPass(), PMW); |
| 183 | |
| 184 | // Do various transformations for exception handling. |
| 185 | // Every CFG-changing optimizations should come before this. |
| 186 | if (TM.Options.ExceptionModel == ExceptionHandling::Wasm) |
| 187 | addMachineFunctionPass(Pass: WebAssemblyLateEHPreparePass(), PMW); |
| 188 | |
| 189 | // Now that we have a prologue and epilogue and all frame indices are |
| 190 | // rewritten, eliminate SP and FP. This allows them to be stackified, |
| 191 | // colored, and numbered with the rest of the registers. |
| 192 | addMachineFunctionPass(Pass: WebAssemblyReplacePhysRegsPass(), PMW); |
| 193 | |
| 194 | // Preparations and optimizations related to register stackification. |
| 195 | if (getOptLevel() != CodeGenOptLevel::None) { |
| 196 | // Depend on LiveIntervals and perform some optimizations on it. |
| 197 | // TODO(boomanaiden154): WebAssemblyOptimizeLiveIntervals |
| 198 | |
| 199 | // Prepare memory intrinsic calls for register stackifying. |
| 200 | // TODO(boomanaiden154): WebAssemblyMemIntrinsicResults |
| 201 | } |
| 202 | |
| 203 | // Mark registers as representing wasm's value stack. This is a key |
| 204 | // code-compression technique in WebAssembly. We run this pass (and |
| 205 | // MemIntrinsicResults above) very late, so that it sees as much code as |
| 206 | // possible, including code emitted by PEI and expanded by late tail |
| 207 | // duplication. |
| 208 | // TODO(boomanaiden154): WebAssemblyRegStackify |
| 209 | |
| 210 | if (getOptLevel() != CodeGenOptLevel::None) { |
| 211 | // Run the register coloring pass to reduce the total number of registers. |
| 212 | // This runs after stackification so that it doesn't consider registers |
| 213 | // that become stackified. |
| 214 | // TODO(boomanaiden154): WebAssemblyRegColoring |
| 215 | } |
| 216 | |
| 217 | // Sort the blocks of the CFG into topological order, a prerequisite for |
| 218 | // BLOCK and LOOP markers. |
| 219 | // TODO(boomanaiden154): WebAssemblyCFGSort |
| 220 | |
| 221 | // Insert BLOCK and LOOP markers. |
| 222 | // TODO(boomanaiden154): WebAssemblyCFGStackify |
| 223 | |
| 224 | // Insert explicit local.get and local.set operators. |
| 225 | if (!WasmDisableExplicitLocals) { |
| 226 | // TODO(boomanaiden154): WebAssemblyExplicitLocals |
| 227 | } |
| 228 | |
| 229 | // Lower br_unless into br_if. |
| 230 | // TODO(boomanaiden154): WebAssemblyLowerBrUnless |
| 231 | |
| 232 | // Perform the very last peephole optimizations on the code. |
| 233 | if (getOptLevel() != CodeGenOptLevel::None) { |
| 234 | // TODO(boomanaiden154): WebAssemblyPeephole |
| 235 | } |
| 236 | |
| 237 | // Create a mapping from LLVM CodeGen virtual registers to wasm registers. |
| 238 | // TODO(boomanaiden154): WebAssemblyRegNumbering |
| 239 | |
| 240 | // Fix debug_values whose defs have been stackified. |
| 241 | if (!WasmDisableExplicitLocals) { |
| 242 | // TODO(boomanaiden154): WebAssemblyDebugFixup |
| 243 | } |
| 244 | |
| 245 | // Collect information to prepare for MC lowering / asm printing. |
| 246 | // TODO(boomanaiden154): WebAssemblyMCLowerPrePass |
| 247 | } |
| 248 | |
| 249 | } // namespace |
| 250 | |
| 251 | void WebAssemblyTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB){ |
| 252 | #define GET_PASS_REGISTRY "WebAssemblyPassRegistry.def" |
| 253 | #include "llvm/Passes/TargetPassRegistry.inc" |
| 254 | } |
| 255 | |
| 256 | Error WebAssemblyTargetMachine::buildCodeGenPipeline( |
| 257 | ModulePassManager &MPM, ModuleAnalysisManager &MAM, raw_pwrite_stream &Out, |
| 258 | raw_pwrite_stream *DwoOut, CodeGenFileType FileType, |
| 259 | const CGPassBuilderOption &Opt, MCContext &Ctx, |
| 260 | PassInstrumentationCallbacks *PIC) { |
| 261 | auto CGPB = WebAssemblyCodeGenPassBuilder(*this, Opt, PIC); |
| 262 | return CGPB.buildPipeline(MPM, MAM, Out, DwoOut, FileType, Ctx); |
| 263 | } |
| 264 | |