stage2: implement `@memset` and `@memcpy` builtins

3 files changed, 117 insertions, 1 deletions

diff --git a/src/codegen/c.zig b/src/codegen/c.zig
index a82f0e57f7..16b13db292 100644
--- a/src/codegen/c.zig
+++ b/src/codegen/c.zig
@@ -953,6 +953,8 @@ fn genBody(f: *Function, body: []const Air.Inst.Index) error{ AnalysisFail, OutO
             .cmpxchg_strong   => try airCmpxchg(f, inst, "strong"),
             .atomic_rmw       => try airAtomicRmw(f, inst),
             .atomic_load      => try airAtomicLoad(f, inst),
+            .memset           => try airMemset(f, inst),
+            .memcpy           => try airMemcpy(f, inst),
 
             .int_to_float,
             .float_to_int,
@@ -2005,8 +2007,12 @@ fn airAtomicRmw(f: *Function, inst: Air.Inst.Index) !CValue {
 
 fn airAtomicLoad(f: *Function, inst: Air.Inst.Index) !CValue {
     const atomic_load = f.air.instructions.items(.data)[inst].atomic_load;
-    const inst_ty = f.air.typeOfIndex(inst);
     const ptr = try f.resolveInst(atomic_load.ptr);
+    const ptr_ty = f.air.typeOf(atomic_load.ptr);
+    if (!ptr_ty.isVolatilePtr() and f.liveness.isUnused(inst))
+        return CValue.none;
+
+    const inst_ty = f.air.typeOfIndex(inst);
     const local = try f.allocLocal(inst_ty, .Const);
     const writer = f.object.writer();
 
@@ -2036,6 +2042,44 @@ fn airAtomicStore(f: *Function, inst: Air.Inst.Index, order: [*:0]const u8) !CVa
     return local;
 }
 
+fn airMemset(f: *Function, inst: Air.Inst.Index) !CValue {
+    const pl_op = f.air.instructions.items(.data)[inst].pl_op;
+    const extra = f.air.extraData(Air.Bin, pl_op.payload).data;
+    const dest_ptr = try f.resolveInst(pl_op.operand);
+    const value = try f.resolveInst(extra.lhs);
+    const len = try f.resolveInst(extra.rhs);
+    const writer = f.object.writer();
+
+    try writer.writeAll("memset(");
+    try f.writeCValue(writer, dest_ptr);
+    try writer.writeAll(", ");
+    try f.writeCValue(writer, value);
+    try writer.writeAll(", ");
+    try f.writeCValue(writer, len);
+    try writer.writeAll(");\n");
+
+    return CValue.none;
+}
+
+fn airMemcpy(f: *Function, inst: Air.Inst.Index) !CValue {
+    const pl_op = f.air.instructions.items(.data)[inst].pl_op;
+    const extra = f.air.extraData(Air.Bin, pl_op.payload).data;
+    const dest_ptr = try f.resolveInst(pl_op.operand);
+    const src_ptr = try f.resolveInst(extra.lhs);
+    const len = try f.resolveInst(extra.rhs);
+    const writer = f.object.writer();
+
+    try writer.writeAll("memcpy(");
+    try f.writeCValue(writer, dest_ptr);
+    try writer.writeAll(", ");
+    try f.writeCValue(writer, src_ptr);
+    try writer.writeAll(", ");
+    try f.writeCValue(writer, len);
+    try writer.writeAll(");\n");
+
+    return CValue.none;
+}
+
 fn toMemoryOrder(order: std.builtin.AtomicOrder) [:0]const u8 {
     return switch (order) {
         .Unordered => "memory_order_relaxed",
diff --git a/src/codegen/llvm.zig b/src/codegen/llvm.zig
index b2b87b97d8..b15834c963 100644
--- a/src/codegen/llvm.zig
+++ b/src/codegen/llvm.zig
@@ -1279,6 +1279,8 @@ pub const FuncGen = struct {
                 .fence          => try self.airFence(inst),
                 .atomic_rmw     => try self.airAtomicRmw(inst),
                 .atomic_load    => try self.airAtomicLoad(inst),
+                .memset         => try self.airMemset(inst),
+                .memcpy         => try self.airMemcpy(inst),
 
                 .atomic_store_unordered => try self.airAtomicStore(inst, .Unordered),
                 .atomic_store_monotonic => try self.airAtomicStore(inst, .Monotonic),
@@ -2426,6 +2428,8 @@ pub const FuncGen = struct {
         const atomic_load = self.air.instructions.items(.data)[inst].atomic_load;
         const ptr = try self.resolveInst(atomic_load.ptr);
         const ptr_ty = self.air.typeOf(atomic_load.ptr);
+        if (!ptr_ty.isVolatilePtr() and self.liveness.isUnused(inst))
+            return null;
         const ordering = toLlvmAtomicOrdering(atomic_load.order);
         const operand_ty = ptr_ty.elemType();
         const opt_abi_ty = self.dg.getAtomicAbiType(operand_ty, false);
@@ -2468,6 +2472,55 @@ pub const FuncGen = struct {
         return null;
     }
 
+    fn airMemset(self: *FuncGen, inst: Air.Inst.Index) !?*const llvm.Value {
+        const pl_op = self.air.instructions.items(.data)[inst].pl_op;
+        const extra = self.air.extraData(Air.Bin, pl_op.payload).data;
+        const dest_ptr = try self.resolveInst(pl_op.operand);
+        const ptr_ty = self.air.typeOf(pl_op.operand);
+        const value = try self.resolveInst(extra.lhs);
+        const val_is_undef = if (self.air.value(extra.lhs)) |val| val.isUndef() else false;
+        const len = try self.resolveInst(extra.rhs);
+        const u8_llvm_ty = self.context.intType(8);
+        const ptr_u8_llvm_ty = u8_llvm_ty.pointerType(0);
+        const dest_ptr_u8 = self.builder.buildBitCast(dest_ptr, ptr_u8_llvm_ty, "");
+        const fill_char = if (val_is_undef) u8_llvm_ty.constInt(0xaa, .False) else value;
+        const target = self.dg.module.getTarget();
+        const dest_ptr_align = ptr_ty.ptrAlignment(target);
+        const memset = self.builder.buildMemSet(dest_ptr_u8, fill_char, len, dest_ptr_align);
+        memset.setVolatile(llvm.Bool.fromBool(ptr_ty.isVolatilePtr()));
+
+        if (val_is_undef and self.dg.module.comp.bin_file.options.valgrind) {
+            // TODO generate valgrind client request to mark byte range as undefined
+            // see gen_valgrind_undef() in codegen.cpp
+        }
+        return null;
+    }
+
+    fn airMemcpy(self: *FuncGen, inst: Air.Inst.Index) !?*const llvm.Value {
+        const pl_op = self.air.instructions.items(.data)[inst].pl_op;
+        const extra = self.air.extraData(Air.Bin, pl_op.payload).data;
+        const dest_ptr = try self.resolveInst(pl_op.operand);
+        const dest_ptr_ty = self.air.typeOf(pl_op.operand);
+        const src_ptr = try self.resolveInst(extra.lhs);
+        const src_ptr_ty = self.air.typeOf(extra.lhs);
+        const len = try self.resolveInst(extra.rhs);
+        const u8_llvm_ty = self.context.intType(8);
+        const ptr_u8_llvm_ty = u8_llvm_ty.pointerType(0);
+        const dest_ptr_u8 = self.builder.buildBitCast(dest_ptr, ptr_u8_llvm_ty, "");
+        const src_ptr_u8 = self.builder.buildBitCast(src_ptr, ptr_u8_llvm_ty, "");
+        const is_volatile = src_ptr_ty.isVolatilePtr() or dest_ptr_ty.isVolatilePtr();
+        const target = self.dg.module.getTarget();
+        const memcpy = self.builder.buildMemCpy(
+            dest_ptr_u8,
+            dest_ptr_ty.ptrAlignment(target),
+            src_ptr_u8,
+            src_ptr_ty.ptrAlignment(target),
+            len,
+        );
+        memcpy.setVolatile(llvm.Bool.fromBool(is_volatile));
+        return null;
+    }
+
     fn getIntrinsic(self: *FuncGen, name: []const u8) *const llvm.Value {
         const id = llvm.lookupIntrinsicID(name.ptr, name.len);
         assert(id != 0);
diff --git a/src/codegen/llvm/bindings.zig b/src/codegen/llvm/bindings.zig
index c53ac08fdd..9d32682260 100644
--- a/src/codegen/llvm/bindings.zig
+++ b/src/codegen/llvm/bindings.zig
@@ -632,6 +632,25 @@ pub const Builder = opaque {
         DestTy: *const Type,
         Name: [*:0]const u8,
     ) *const Value;
+
+    pub const buildMemSet = LLVMBuildMemSet;
+    extern fn LLVMBuildMemSet(
+        B: *const Builder,
+        Ptr: *const Value,
+        Val: *const Value,
+        Len: *const Value,
+        Align: c_uint,
+    ) *const Value;
+
+    pub const buildMemCpy = LLVMBuildMemCpy;
+    extern fn LLVMBuildMemCpy(
+        B: *const Builder,
+        Dst: *const Value,
+        DstAlign: c_uint,
+        Src: *const Value,
+        SrcAlign: c_uint,
+        Size: *const Value,
+    ) *const Value;
 };
 
 pub const IntPredicate = enum(c_uint) {