better handling of arrays in packed structs

* Separate LoadPtr IR instructions into pass1 and pass2 variants.
 * Define `type_size_bits` for extern structs to be the same as
   their `@sizeOf(T) * 8` and allow them in packed structs.
 * More helpful error messages when trying to use types in
   packed structs that are not allowed.
 * Support arrays in packed structs even when they are not
   byte-aligned.
 * Add compile error for using arrays in packed structs when the
   padding bits would be problematic. This is necessary since
   we do not have packed arrays.

closes #677
This commit is contained in:
Andrew Kelley
2019-02-21 14:44:14 -05:00
parent 2bb795dc45
commit 1066004b79
7 changed files with 248 additions and 52 deletions

View File

@@ -3281,7 +3281,7 @@ static LLVMValueRef ir_render_decl_var(CodeGen *g, IrExecutable *executable,
return nullptr;
}
static LLVMValueRef ir_render_load_ptr(CodeGen *g, IrExecutable *executable, IrInstructionLoadPtr *instruction) {
static LLVMValueRef ir_render_load_ptr(CodeGen *g, IrExecutable *executable, IrInstructionLoadPtrGen *instruction) {
ZigType *child_type = instruction->base.value.type;
if (!type_has_bits(child_type))
return nullptr;
@@ -3296,7 +3296,6 @@ static LLVMValueRef ir_render_load_ptr(CodeGen *g, IrExecutable *executable, IrI
bool big_endian = g->is_big_endian;
assert(!handle_is_ptr(child_type));
LLVMValueRef containing_int = gen_load(g, ptr, ptr_type, "");
uint32_t host_bit_count = LLVMGetIntTypeWidth(LLVMTypeOf(containing_int));
assert(host_bit_count == host_int_bytes * 8);
@@ -3308,7 +3307,16 @@ static LLVMValueRef ir_render_load_ptr(CodeGen *g, IrExecutable *executable, IrI
LLVMValueRef shift_amt_val = LLVMConstInt(LLVMTypeOf(containing_int), shift_amt, false);
LLVMValueRef shifted_value = LLVMBuildLShr(g->builder, containing_int, shift_amt_val, "");
return LLVMBuildTrunc(g->builder, shifted_value, child_type->type_ref, "");
if (!handle_is_ptr(child_type))
return LLVMBuildTrunc(g->builder, shifted_value, child_type->type_ref, "");
assert(instruction->tmp_ptr != nullptr);
LLVMTypeRef same_size_int = LLVMIntType(size_in_bits);
LLVMValueRef truncated_int = LLVMBuildTrunc(g->builder, shifted_value, same_size_int, "");
LLVMValueRef bitcasted_ptr = LLVMBuildBitCast(g->builder, instruction->tmp_ptr,
LLVMPointerType(same_size_int, 0), "");
LLVMBuildStore(g->builder, truncated_int, bitcasted_ptr);
return instruction->tmp_ptr;
}
static bool value_is_all_undef_array(ConstExprValue *const_val, size_t len) {
@@ -5460,6 +5468,7 @@ static LLVMValueRef ir_render_instruction(CodeGen *g, IrExecutable *executable,
case IrInstructionIdDeclVarSrc:
case IrInstructionIdPtrCastSrc:
case IrInstructionIdCmpxchgSrc:
case IrInstructionIdLoadPtr:
zig_unreachable();
case IrInstructionIdDeclVarGen:
@@ -5478,8 +5487,8 @@ static LLVMValueRef ir_render_instruction(CodeGen *g, IrExecutable *executable,
return ir_render_br(g, executable, (IrInstructionBr *)instruction);
case IrInstructionIdUnOp:
return ir_render_un_op(g, executable, (IrInstructionUnOp *)instruction);
case IrInstructionIdLoadPtr:
return ir_render_load_ptr(g, executable, (IrInstructionLoadPtr *)instruction);
case IrInstructionIdLoadPtrGen:
return ir_render_load_ptr(g, executable, (IrInstructionLoadPtrGen *)instruction);
case IrInstructionIdStorePtr:
return ir_render_store_ptr(g, executable, (IrInstructionStorePtr *)instruction);
case IrInstructionIdVarPtr:
@@ -5836,8 +5845,32 @@ static LLVMValueRef pack_const_int(CodeGen *g, LLVMTypeRef big_int_type_ref, Con
LLVMValueRef ptr_size_int_val = LLVMConstPtrToInt(ptr_val, g->builtin_types.entry_usize->type_ref);
return LLVMConstZExt(ptr_size_int_val, big_int_type_ref);
}
case ZigTypeIdArray:
zig_panic("TODO bit pack an array");
case ZigTypeIdArray: {
LLVMValueRef val = LLVMConstInt(big_int_type_ref, 0, false);
if (const_val->data.x_array.special == ConstArraySpecialUndef) {
return val;
}
expand_undef_array(g, const_val);
bool is_big_endian = g->is_big_endian; // TODO get endianness from struct type
uint32_t packed_bits_size = type_size_bits(g, type_entry->data.array.child_type);
size_t used_bits = 0;
for (size_t i = 0; i < type_entry->data.array.len; i += 1) {
ConstExprValue *elem_val = &const_val->data.x_array.data.s_none.elements[i];
LLVMValueRef child_val = pack_const_int(g, big_int_type_ref, elem_val);
if (is_big_endian) {
LLVMValueRef shift_amt = LLVMConstInt(big_int_type_ref, packed_bits_size, false);
val = LLVMConstShl(val, shift_amt);
val = LLVMConstOr(val, child_val);
} else {
LLVMValueRef shift_amt = LLVMConstInt(big_int_type_ref, used_bits, false);
LLVMValueRef child_val_shifted = LLVMConstShl(child_val, shift_amt);
val = LLVMConstOr(val, child_val_shifted);
used_bits += packed_bits_size;
}
}
return val;
}
case ZigTypeIdVector:
zig_panic("TODO bit pack a vector");
case ZigTypeIdUnion:
@@ -6728,6 +6761,9 @@ static void do_code_gen(CodeGen *g) {
} else if (instruction->id == IrInstructionIdResizeSlice) {
IrInstructionResizeSlice *resize_slice_instruction = (IrInstructionResizeSlice *)instruction;
slot = &resize_slice_instruction->tmp_ptr;
} else if (instruction->id == IrInstructionIdLoadPtrGen) {
IrInstructionLoadPtrGen *load_ptr_inst = (IrInstructionLoadPtrGen *)instruction;
slot = &load_ptr_inst->tmp_ptr;
} else if (instruction->id == IrInstructionIdVectorToArray) {
IrInstructionVectorToArray *vector_to_array_instruction = (IrInstructionVectorToArray *)instruction;
alignment_bytes = get_abi_alignment(g, vector_to_array_instruction->vector->value.type);