codegen: define return_roots in normalized form (#39745)

LLVM will switch to this form, so it is preferable to start that way. Refs Matcha in #39641 (cherry picked from commit f879cd1)
JuliaLang · Dec 22, 2022 · 7d46071 · 7d46071
1 parent 8071f26
commit 7d46071
Show file tree

Hide file tree

Showing 2 changed files with 13 additions and 13 deletions.
diff --git a/src/codegen.cpp b/src/codegen.cpp
@@ -3307,8 +3307,7 @@ static jl_cgval_t emit_call_specfun_other(jl_codectx_t &ctx, jl_method_instance_
     }
 
     if (returninfo.return_roots) {
-        AllocaInst *return_roots = emit_static_alloca(ctx, T_prjlvalue);
-        return_roots->setOperand(0, ConstantInt::get(T_int32, returninfo.return_roots));
+        AllocaInst *return_roots = emit_static_alloca(ctx, ArrayType::get(T_prjlvalue, returninfo.return_roots));
         argvals[idx] = return_roots;
         idx++;
     }
@@ -4708,8 +4707,11 @@ static void emit_cfunc_invalidate(
         break;
     }
     case jl_returninfo_t::SRet: {
-        if (return_roots)
-            ctx.builder.CreateStore(gf_ret, gf_thunk->arg_begin() + 1);
+        if (return_roots) {
+            Value *root1 = gf_thunk->arg_begin() + 1; // root1 has type [n x {}*]*
+            root1 = ctx.builder.CreateConstInBoundsGEP2_32(root1->getType()->getPointerElementType(), root1, 0, 0);
+            ctx.builder.CreateStore(gf_ret, root1);
+        }
         emit_memcpy(ctx, &*gf_thunk->arg_begin(), nullptr, gf_ret, nullptr, jl_datatype_size(rettype), julia_alignment(rettype));
         ctx.builder.CreateRetVoid();
         break;
@@ -5085,8 +5087,7 @@ static Function* gen_cfun_wrapper(
             args.push_back(result);
         }
         if (returninfo.return_roots) {
-            AllocaInst *return_roots = emit_static_alloca(ctx, T_prjlvalue);
-            return_roots->setOperand(0, ConstantInt::get(T_int32, returninfo.return_roots));
+            AllocaInst *return_roots = emit_static_alloca(ctx, ArrayType::get(T_prjlvalue, returninfo.return_roots));
             args.push_back(return_roots);
         }
         for (size_t i = 0; i < nargs + 1; i++) {
@@ -5506,8 +5507,7 @@ static Function *gen_invoke_wrapper(jl_method_instance_t *lam, jl_value_t *jlret
         break;
     }
     if (f.return_roots) {
-        AllocaInst *return_roots = emit_static_alloca(ctx, T_prjlvalue);
-        return_roots->setOperand(0, ConstantInt::get(T_int32, f.return_roots));
+        AllocaInst *return_roots = emit_static_alloca(ctx, ArrayType::get(T_prjlvalue, f.return_roots));
         args[idx] = return_roots;
         idx++;
     }
@@ -5654,7 +5654,7 @@ static jl_returninfo_t get_specsig_function(jl_codectx_t &ctx, Module *M, String
     }
 
     if (props.return_roots) {
-        fsig.push_back(T_pprjlvalue);
+        fsig.push_back(ArrayType::get(T_prjlvalue, props.return_roots)->getPointerTo(0));
         unsigned argno = fsig.size();
         attributes = attributes.addAttribute(jl_LLVMContext, argno, Attribute::NoAlias);
         attributes = attributes.addAttribute(jl_LLVMContext, argno, Attribute::NoCapture);

diff --git a/src/llvm-late-gc-lowering.cpp b/src/llvm-late-gc-lowering.cpp
@@ -1762,8 +1762,8 @@ unsigned TrackWithShadow(Value *Src, Type *STy, bool isptr, Value *Dst, IRBuilde
     auto Ptrs = ExtractTrackedValues(Src, STy, isptr, irbuilder);
     for (unsigned i = 0; i < Ptrs.size(); ++i) {
         Value *Elem = Ptrs[i];
-        assert(Elem->getType()->isPointerTy());
-        Value *Slot = irbuilder.CreateConstInBoundsGEP1_32(Elem->getType(), Dst, i);
+        Type *ET = Dst->getType()->getPointerElementType(); // Dst has type `[n x {}*]*`
+        Value *Slot = irbuilder.CreateConstInBoundsGEP2_32(ET, Dst, 0, i);
         StoreInst *shadowStore = irbuilder.CreateAlignedStore(Elem, Slot, Align(sizeof(void*)));
         shadowStore->setOrdering(AtomicOrdering::NotAtomic);
         // TODO: shadowStore->setMetadata(LLVMContext::MD_tbaa, tbaa_gcframe);
@@ -1790,7 +1790,7 @@ void LateLowerGCFrame::MaybeTrackDst(State &S, MemTransferInst *MI) {
     //            Src = new BitCastInst(Src, STy->getPointerTo(MI->getSourceAddressSpace()), "", MI);
     //            auto &Shadow = S.ShadowAllocas[AI];
     //            if (!Shadow)
-    //                Shadow = new AllocaInst(T_prjlvalue, 0, ConstantInt::get(T_int32, nroots), "", MI);
+    //                Shadow = new AllocaInst(ArrayType::get(T_prjlvalue, nroots), 0, "", MI);
     //            AI = Shadow;
     //            unsigned count = TrackWithShadow(Src, STy, true, AI, IRBuilder<>(MI));
     //            assert(count == tracked.count); (void)count;
@@ -1826,7 +1826,7 @@ void LateLowerGCFrame::MaybeTrackStore(State &S, StoreInst *I) {
     // track the Store with a Shadow
     //auto &Shadow = S.ShadowAllocas[AI];
     //if (!Shadow)
-    //    Shadow = new AllocaInst(T_prjlvalue, 0, ConstantInt::get(T_int32, tracked.count), "", MI);
+    //    Shadow = new AllocaInst(ArrayType::get(T_prjlvalue, tracked.count), 0, "", MI);
     //AI = Shadow;
     //Value *Src = I->getValueOperand();
     //unsigned count = TrackWithShadow(Src, Src->getType(), false, AI, MI, TODO which slots are we actually clobbering?);