aboutsummaryrefslogtreecommitdiff
path: root/clang/lib/CodeGen/TargetBuiltins
diff options
context:
space:
mode:
Diffstat (limited to 'clang/lib/CodeGen/TargetBuiltins')
-rw-r--r--clang/lib/CodeGen/TargetBuiltins/AMDGPU.cpp42
-rw-r--r--clang/lib/CodeGen/TargetBuiltins/ARM.cpp4
-rw-r--r--clang/lib/CodeGen/TargetBuiltins/WebAssembly.cpp61
3 files changed, 104 insertions, 3 deletions
diff --git a/clang/lib/CodeGen/TargetBuiltins/AMDGPU.cpp b/clang/lib/CodeGen/TargetBuiltins/AMDGPU.cpp
index ee736a2..70f510a 100644
--- a/clang/lib/CodeGen/TargetBuiltins/AMDGPU.cpp
+++ b/clang/lib/CodeGen/TargetBuiltins/AMDGPU.cpp
@@ -274,7 +274,7 @@ llvm::Value *CodeGenFunction::EmitScalarOrConstFoldImmArg(unsigned ICEArguments,
void CodeGenFunction::AddAMDGPUFenceAddressSpaceMMRA(llvm::Instruction *Inst,
const CallExpr *E) {
- constexpr const char *Tag = "amdgpu-as";
+ constexpr const char *Tag = "amdgpu-synchronize-as";
LLVMContext &Ctx = Inst->getContext();
SmallVector<MMRAMetadata::TagT, 3> MMRAs;
@@ -633,6 +633,41 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
llvm::Function *F = CGM.getIntrinsic(IID, {LoadTy});
return Builder.CreateCall(F, {Addr});
}
+ case AMDGPU::BI__builtin_amdgcn_global_load_monitor_b32:
+ case AMDGPU::BI__builtin_amdgcn_global_load_monitor_b64:
+ case AMDGPU::BI__builtin_amdgcn_global_load_monitor_b128:
+ case AMDGPU::BI__builtin_amdgcn_flat_load_monitor_b32:
+ case AMDGPU::BI__builtin_amdgcn_flat_load_monitor_b64:
+ case AMDGPU::BI__builtin_amdgcn_flat_load_monitor_b128: {
+
+ Intrinsic::ID IID;
+ switch (BuiltinID) {
+ case AMDGPU::BI__builtin_amdgcn_global_load_monitor_b32:
+ IID = Intrinsic::amdgcn_global_load_monitor_b32;
+ break;
+ case AMDGPU::BI__builtin_amdgcn_global_load_monitor_b64:
+ IID = Intrinsic::amdgcn_global_load_monitor_b64;
+ break;
+ case AMDGPU::BI__builtin_amdgcn_global_load_monitor_b128:
+ IID = Intrinsic::amdgcn_global_load_monitor_b128;
+ break;
+ case AMDGPU::BI__builtin_amdgcn_flat_load_monitor_b32:
+ IID = Intrinsic::amdgcn_flat_load_monitor_b32;
+ break;
+ case AMDGPU::BI__builtin_amdgcn_flat_load_monitor_b64:
+ IID = Intrinsic::amdgcn_flat_load_monitor_b64;
+ break;
+ case AMDGPU::BI__builtin_amdgcn_flat_load_monitor_b128:
+ IID = Intrinsic::amdgcn_flat_load_monitor_b128;
+ break;
+ }
+
+ llvm::Type *LoadTy = ConvertType(E->getType());
+ llvm::Value *Addr = EmitScalarExpr(E->getArg(0));
+ llvm::Value *Val = EmitScalarExpr(E->getArg(1));
+ llvm::Function *F = CGM.getIntrinsic(IID, {LoadTy});
+ return Builder.CreateCall(F, {Addr, Val});
+ }
case AMDGPU::BI__builtin_amdgcn_load_to_lds: {
// Should this have asan instrumentation?
return emitBuiltinWithOneOverloadedType<5>(*this, E,
@@ -855,6 +890,7 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
case AMDGPU::BI__builtin_amdgcn_wmma_f32_16x16x128_bf8_fp8:
case AMDGPU::BI__builtin_amdgcn_wmma_f32_16x16x128_bf8_bf8:
case AMDGPU::BI__builtin_amdgcn_wmma_i32_16x16x64_iu8:
+ case AMDGPU::BI__builtin_amdgcn_wmma_f32_16x16x128_f8f6f4:
case AMDGPU::BI__builtin_amdgcn_wmma_f32_32x16x128_f4:
case AMDGPU::BI__builtin_amdgcn_swmmac_f32_16x16x64_f16:
case AMDGPU::BI__builtin_amdgcn_swmmac_f32_16x16x64_bf16:
@@ -1118,6 +1154,10 @@ Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
ArgsForMatchingMatrixTypes = {4, 1};
BuiltinWMMAOp = Intrinsic::amdgcn_wmma_i32_16x16x64_iu8;
break;
+ case AMDGPU::BI__builtin_amdgcn_wmma_f32_16x16x128_f8f6f4:
+ ArgsForMatchingMatrixTypes = {5, 1, 3};
+ BuiltinWMMAOp = Intrinsic::amdgcn_wmma_f32_16x16x128_f8f6f4;
+ break;
case AMDGPU::BI__builtin_amdgcn_wmma_f32_32x16x128_f4:
ArgsForMatchingMatrixTypes = {3, 0, 1};
BuiltinWMMAOp = Intrinsic::amdgcn_wmma_f32_32x16x128_f4;
diff --git a/clang/lib/CodeGen/TargetBuiltins/ARM.cpp b/clang/lib/CodeGen/TargetBuiltins/ARM.cpp
index 7e6a47f..2e6b4b3 100644
--- a/clang/lib/CodeGen/TargetBuiltins/ARM.cpp
+++ b/clang/lib/CodeGen/TargetBuiltins/ARM.cpp
@@ -8112,7 +8112,7 @@ Value *CodeGenFunction::EmitAArch64CpuSupports(const CallExpr *E) {
llvm::Value *
CodeGenFunction::EmitAArch64CpuSupports(ArrayRef<StringRef> FeaturesStrs) {
- uint64_t FeaturesMask = llvm::AArch64::getCpuSupportsMask(FeaturesStrs);
+ llvm::APInt FeaturesMask = llvm::AArch64::getCpuSupportsMask(FeaturesStrs);
Value *Result = Builder.getTrue();
if (FeaturesMask != 0) {
// Get features from structure in runtime library
@@ -8128,7 +8128,7 @@ CodeGenFunction::EmitAArch64CpuSupports(ArrayRef<StringRef> FeaturesStrs) {
{ConstantInt::get(Int32Ty, 0), ConstantInt::get(Int32Ty, 0)});
Value *Features = Builder.CreateAlignedLoad(Int64Ty, CpuFeatures,
CharUnits::fromQuantity(8));
- Value *Mask = Builder.getInt64(FeaturesMask);
+ Value *Mask = Builder.getInt(FeaturesMask.trunc(64));
Value *Bitset = Builder.CreateAnd(Features, Mask);
Value *Cmp = Builder.CreateICmpEQ(Bitset, Mask);
Result = Builder.CreateAnd(Result, Cmp);
diff --git a/clang/lib/CodeGen/TargetBuiltins/WebAssembly.cpp b/clang/lib/CodeGen/TargetBuiltins/WebAssembly.cpp
index b7fd70e..33a8d8f 100644
--- a/clang/lib/CodeGen/TargetBuiltins/WebAssembly.cpp
+++ b/clang/lib/CodeGen/TargetBuiltins/WebAssembly.cpp
@@ -12,7 +12,10 @@
#include "CGBuiltin.h"
#include "clang/Basic/TargetBuiltins.h"
+#include "llvm/ADT/APInt.h"
+#include "llvm/IR/Constants.h"
#include "llvm/IR/IntrinsicsWebAssembly.h"
+#include "llvm/Support/ErrorHandling.h"
using namespace clang;
using namespace CodeGen;
@@ -218,6 +221,64 @@ Value *CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_ref_null_func);
return Builder.CreateCall(Callee);
}
+ case WebAssembly::BI__builtin_wasm_test_function_pointer_signature: {
+ Value *FuncRef = EmitScalarExpr(E->getArg(0));
+
+ // Get the function type from the argument's static type
+ QualType ArgType = E->getArg(0)->getType();
+ const PointerType *PtrTy = ArgType->getAs<PointerType>();
+ assert(PtrTy && "Sema should have ensured this is a function pointer");
+
+ const FunctionType *FuncTy = PtrTy->getPointeeType()->getAs<FunctionType>();
+ assert(FuncTy && "Sema should have ensured this is a function pointer");
+
+ // In the llvm IR, we won't have access any more to the type of the function
+ // pointer so we need to insert this type information somehow. The
+ // @llvm.wasm.ref.test.func takes varargs arguments whose values are unused
+ // to indicate the type of the function to test for. See the test here:
+ // llvm/test/CodeGen/WebAssembly/ref-test-func.ll
+ //
+ // The format is: first we include the return types (since this is a C
+ // function pointer, there will be 0 or one of these) then a token type to
+ // indicate the boundary between return types and param types, then the
+ // param types.
+
+ llvm::FunctionType *LLVMFuncTy =
+ cast<llvm::FunctionType>(ConvertType(QualType(FuncTy, 0)));
+
+ unsigned NParams = LLVMFuncTy->getNumParams();
+ std::vector<Value *> Args;
+ Args.reserve(NParams + 3);
+ // The only real argument is the FuncRef
+ Args.push_back(FuncRef);
+
+ // Add the type information
+ auto addType = [this, &Args](llvm::Type *T) {
+ if (T->isVoidTy()) {
+ // Do nothing
+ } else if (T->isFloatingPointTy()) {
+ Args.push_back(ConstantFP::get(T, 0));
+ } else if (T->isIntegerTy()) {
+ Args.push_back(ConstantInt::get(T, 0));
+ } else if (T->isPointerTy()) {
+ Args.push_back(ConstantPointerNull::get(llvm::PointerType::get(
+ getLLVMContext(), T->getPointerAddressSpace())));
+ } else {
+ // TODO: Handle reference types. For now, we reject them in Sema.
+ llvm_unreachable("Unhandled type");
+ }
+ };
+
+ addType(LLVMFuncTy->getReturnType());
+ // The token type indicates the boundary between return types and param
+ // types.
+ Args.push_back(PoisonValue::get(llvm::Type::getTokenTy(getLLVMContext())));
+ for (unsigned i = 0; i < NParams; i++) {
+ addType(LLVMFuncTy->getParamType(i));
+ }
+ Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_ref_test_func);
+ return Builder.CreateCall(Callee, Args);
+ }
case WebAssembly::BI__builtin_wasm_swizzle_i8x16: {
Value *Src = EmitScalarExpr(E->getArg(0));
Value *Indices = EmitScalarExpr(E->getArg(1));
generated by cgit v1.1 at 2025-07-28 12:18:23 +0000