c3c/src/compiler/sema_asm.c

// Copyright (c) 2022-2024 Christoffer Lerno and contributors. All rights reserved.
// Use of this source code is governed by a LGPLv3.0
// a copy of which can be found in the LICENSE file.

#include "sema_internal.h"
#include "compiler/asm/x86.h"

// Add a single clobber to a block.
static inline void sema_add_clobber(AsmInlineBlock *block, unsigned index)
{
	clobbers_add(&block->clobbers, index);
}

// Add a full clobber mask to the clobbers.
static inline void sema_add_clobbers(AsmInlineBlock *block, Clobbers *clobbers)
{
	for (unsigned i = 0; i < CLOBBER_FLAG_ELEMENTS; i++)
	{
		block->clobbers.mask[i] |= clobbers->mask[i];
	}
}

static inline Type *max_supported_imm_int(bool is_signed, AsmArgType arg)
{
	// We don't support i128 as imm ints as no targets do.
	if (is_signed)
	{
		unsigned bits = arg_bits_max(arg.imm_arg_ibits, 64);
		if (!bits)
		{
			bits = arg_bits_max(arg.imm_arg_ubits, 64);
			if (!bits) return NULL;
			return type_int_unsigned_by_bitsize(next_highest_power_of_2(bits));
		}
		return type_int_signed_by_bitsize(next_highest_power_of_2(bits));
	}
	unsigned bits = arg_bits_max(arg.imm_arg_ubits, 64);
	if (!bits) return NULL;
	return type_int_unsigned_by_bitsize(next_highest_power_of_2(bits));
}

/*
static inline AsmArgGroup sema_ireg_for_type(Type *type)
{
	switch (type_size(type))
	{
		case 1:
			return AARG_R8;
		case 2:
			return AARG_R16;
		case 4:
			return AARG_R32;
		case 8:
			return AARG_R64;
		case 16:
			return AARG_R128;
		default:
			UNREACHABLE
	}
}
 */
static inline bool sema_reg_int_suported_type(AsmArgType arg, Type *type)
{
	ASSERT0(type_flatten(type) == type);
	unsigned bits = type_bit_size(type);
	return next_highest_power_of_2(arg_bits_max(arg.ireg_bits, bits)) == bits;
}

INLINE bool sema_reg_is_valid_in_slot(AsmRegister *reg, AsmArgType arg_type)
{
	switch (reg->type)
	{
		case ASM_REG_INT:
			return (arg_type.ireg_bits & reg->bits) != 0;
		case ASM_REG_FLOAT:
			return (arg_type.float_bits & reg->bits) != 0;
		case ASM_REF_FVEC:
		case ASM_REG_IVEC:
			return (arg_type.vec_bits & reg->bits) != 0;
	}
	UNREACHABLE
}

static inline bool sema_reg_float_suported_type(AsmArgType arg, Type *type)
{
	ASSERT0(type_flatten(type) == type);
	if (!arg.float_bits) return false;
	return type_bit_size(type) == next_highest_power_of_2(arg_bits_max(arg.float_bits, 0));
}

static inline bool sema_check_npot_imm_fits(Int imm, AsmArgType arg_type)
{
	// Check if actually an immediate. If not, just move along.
	if (arg_type.imm_arg_ibits == 0 && arg_type.imm_arg_ubits == 0) return true;
	// See if we can do a direct comparison
	bool direct_compare = int_fits(imm, TYPE_I64);
	int64_t val = direct_compare ? int_to_i64(imm) : 0;

	// The signed case
	if (arg_type.imm_arg_ibits > 0)
	{
		if (arg_type.imm_arg_ibits & ARG_BITS_20)
		{
			if (!direct_compare) return false;
			return val >= INT20_MIN && val <= INT20_MAX;
		}
		if (arg_type.imm_arg_ibits & ARG_BITS_12)
		{
			if (!direct_compare) return false;
			return val >= INT12_MIN && val <= INT12_MAX;
		}
		if (arg_type.imm_arg_ibits & ARG_BITS_5)
		{
			if (!direct_compare) return false;
			return val >= INT5_MIN && val <= INT5_MAX;
		}
		return true;
	}
	ASSERT0(arg_type.imm_arg_ubits > 0);
	if (arg_type.imm_arg_ubits & ARG_BITS_20)
	{
		if (!direct_compare) return false;
		return val >= 0 && val <= UINT20_MAX;
	}
	if (arg_type.imm_arg_ubits & ARG_BITS_12)
	{
		if (!direct_compare) return false;
		return val >= 0 && val <= UINT12_MAX;
	}
	if (arg_type.imm_arg_ubits & ARG_BITS_5)
	{
		if (!direct_compare) return false;
		return val >= 0 && val <= UINT5_MAX;
	}
	return true;
}

static inline bool sema_check_asm_arg_const_int(SemaContext *context, AsmInlineBlock *block, AsmInstruction *instr, AsmArgType arg_type, Expr *expr, Expr *int_expr)
{
	bool is_signed = type_kind_is_signed(int_expr->const_expr.ixx.type);
	Type *type = max_supported_imm_int(is_signed, arg_type);
	if (!type)
	{
		SEMA_ERROR(expr, "'%s' does not support a direct integer constant here.", instr->name);
		return false;
	}
	Int i = int_expr->const_expr.ixx;
	unsigned max_bits = arg_bits_max(arg_type.imm_arg_ibits > arg_type.imm_arg_ubits ? arg_type.imm_arg_ibits : arg_type.imm_arg_ubits, 0);
	if (!type || !int_fits(i, type->type_kind) || !sema_check_npot_imm_fits(i, arg_type))
	{
		SEMA_ERROR(expr, "'%s' expected %s limited to %d bits.", instr->name, type_quoted_error_string(type), max_bits);
		return false;
	}
	// Because we assume max 64 bit imm, we can do this simple cast for signed values.
	expr->expr_asm_arg.is_neg = false;
	expr->expr_asm_arg.bits = max_bits;
	if (is_signed)
	{
		expr->expr_asm_arg.is_neg = i128_is_neg(int_expr->const_expr.ixx.i);
		switch (type->type_kind)
		{
			case TYPE_I8:
				i.i.low &= 0xFF;
				break;
			case TYPE_I16:
				i.i.low &= 0xFFFF;
				break;
			case TYPE_I32:
				i.i.low &= 0xFFFFFFFF;
				break;
			default:
				break;
		}
	}
	expr->expr_asm_arg.kind = ASM_ARG_INT;
	expr->expr_asm_arg.value = i.i.low;
	expr->type = type;
	return true;
}

static inline bool sema_check_asm_arg(SemaContext *context, AsmInlineBlock *block, AsmInstruction *instr, AsmArgType arg_type, Expr *expr);

static inline bool sema_check_asm_arg_addr(SemaContext *context, AsmInlineBlock *block, AsmInstruction *instr, AsmArgType arg_type, Expr *expr)
{
	if (!arg_type.is_address)
	{
		RETURN_SEMA_ERROR(expr, "An address cannot appear in this slot.");
	}
	ExprAsmArg *asm_arg = &expr->expr_asm_arg;
	Expr *base = exprptr(asm_arg->base);
	ASSERT0(base->expr_kind == EXPR_ASM);
	ExprAsmArg *base_arg = &base->expr_asm_arg;
	AsmArgType any_ireg = { .ireg_bits = (AsmArgBits)0xFF };
	unsigned bit_size = 0;
	switch (base_arg->kind)
	{
		case ASM_ARG_REG:
			if (!sema_check_asm_arg(context, block, instr, any_ireg, base)) return false;
			bit_size = arg_bits_max(base_arg->reg.ref->bits, 0);
			break;
		case ASM_ARG_REGVAR:
			if (!sema_check_asm_arg(context, block, instr, any_ireg, base)) return false;
			bit_size = type_bit_size(base_arg->ident.ident_decl->type);
			break;
		case ASM_ARG_ADDROF:
			TODO
			break;
		default:
			RETURN_SEMA_ERROR(expr, "Expected a register here.");
	}
	Expr *index = exprptrzero(asm_arg->idx);

	if (index)
	{
		unsigned index_size = 0;
		ExprAsmArg *index_arg = &index->expr_asm_arg;
		switch (index_arg->kind)
		{
			case ASM_ARG_REG:
				if (!sema_check_asm_arg(context, block, instr, any_ireg, index)) return false;
				index_size = arg_bits_max(base_arg->reg.ref->bits, 0);
				break;
			case ASM_ARG_REGVAR:
				if (!sema_check_asm_arg(context, block, instr, any_ireg, index)) return false;
				index_size = type_bit_size(index_arg->ident.ident_decl->type);
				break;
			default:
				SEMA_ERROR(expr, "Expected a register here.");
				return false;
		}
		if (bit_size != index_size)
		{
			RETURN_SEMA_ERROR(index, "Expected the same register size as for the base value.");
		}
	}
	if ((compiler.platform.arch == ARCH_TYPE_RISCV32 || 
		compiler.platform.arch == ARCH_TYPE_RISCV64) &&
		asm_arg->offset)
	{
		if ((asm_arg->neg_offset && asm_arg->offset > abs(INT12_MIN)) ||
			(!asm_arg->neg_offset && asm_arg->offset > INT12_MAX))
		{
			RETURN_SEMA_ERROR(expr, "RISC-V offset limited to 12-bits signed.");
		}
	}

	REMINDER("check if addressing mode is supported");
	return true;
}

static inline bool sema_check_asm_arg_reg(SemaContext *context, AsmInlineBlock *block, AsmInstruction *instr, AsmArgType arg_type, Expr *expr)
{
	const char *name = expr->expr_asm_arg.reg.name;
	AsmRegister *reg = expr->expr_asm_arg.reg.ref = asm_reg_by_name(&compiler.platform, name);
	if (!reg) RETURN_SEMA_ERROR(expr, "Expected a valid register name.");
	if (!sema_reg_is_valid_in_slot(reg, arg_type))
	{
		RETURN_SEMA_ERROR(expr, "'%s' is not valid in this slot.", reg->name);
	}
	if (arg_type.is_write)
	{
		sema_add_clobber(block, reg->clobber_index);
	}
	return true;
}

static inline ExprAsmArg *asm_reg_find_decl(ExprAsmArg **args, Decl *decl, AsmArgKind kind)
{
	FOREACH(ExprAsmArg *, val, args)
	{
		if (val->kind == kind && val->ident.ident_decl == decl) return val;
	}
	return NULL;

}
static inline void asm_reg_add_output(AsmInlineBlock *block, ExprAsmArg *arg)
{
	Decl *decl = arg->ident.ident_decl;

	// Check if this is already in the outputs
	ExprAsmArg *out = asm_reg_find_decl(block->output_vars, decl, arg->kind);
	if (out)
	{
		// Just copy
		arg->ident = out->ident;
	}
	else
	{
		// Add a new
		unsigned out_count = vec_size(block->output_vars);
		if (out_count > 0xFFFF) error_exit("Too many output vars.");
		arg->index = out_count;
		vec_add(block->output_vars, arg);

		// Ignore for memvar
		if (arg->kind == ASM_ARG_MEMVAR) return;

		// It might be in the inputs
		ExprAsmArg *in = asm_reg_find_decl(block->input, decl, arg->kind);
		if (in)
		{
			// It is, so make the in a copy.
			in->ident.copy_output = true;
			in->index = arg->index;
		}
	}

}

static inline void asm_add_input(AsmInlineBlock *block, ExprAsmArg *arg)
{
	unsigned in_count = vec_size(block->input);
	if (in_count > 0xFFFF) error_exit("Too many input vars.");
	arg->index = in_count;
	vec_add(block->input, arg);
}

static inline void asm_reg_add_input(AsmInlineBlock *block, ExprAsmArg *arg)
{
	Decl *decl = arg->ident.ident_decl;

	// 1. It might be an output var, if so then copy the index.
	ExprAsmArg *out = arg->kind == ASM_ARG_MEMVAR ? NULL : asm_reg_find_decl(block->output_vars, decl, arg->kind);
	if (out)
	{
		arg->ident.copy_output = true;
		out->ident.copy_output = true;
		arg->index = out->index;
		goto ADD_CLOBBER;
	}

	// 2. Look through existing inputs, if it's found then copy the index.
	ExprAsmArg *in = asm_reg_find_decl(block->input, decl, arg->kind);
	if (in)
	{
		arg->index = in->index;
	}
	else
	{
		asm_add_input(block, arg);
	}
ADD_CLOBBER:;
	FOREACH(ExprAsmArg *, val, block->output_vars) val->ident.early_clobber = true;
}

static inline bool sema_check_asm_var(SemaContext *context, AsmInlineBlock *block, AsmInstruction *instr, AsmArgType arg_type, Expr *expr)
{
	ExprAsmArg *arg = &expr->expr_asm_arg;
	const char *name = arg->ident.name;
	Decl *decl = sema_resolve_symbol(context, name, NULL, expr->span);
	if (!decl) return false;

	ASSERT0(arg->kind == ASM_ARG_REGVAR);
	arg->ident.ident_decl = decl;
	if (decl->decl_kind != DECL_VAR)
	{
		SEMA_ERROR(expr, "Expected a global or local variable.");
		return false;
	}
	if (IS_OPTIONAL(decl))
	{
		SEMA_ERROR(expr, "Optional variables are not allowed in asm.");
		return false;
	}
	bool is_write = arg_type.is_write;
	bool is_read = !arg_type.is_write || arg_type.is_readwrite;
	arg->ident.is_input = !is_write;
	if (is_read)
	{
		decl->var.is_read = true;
		if (decl->var.out_param)
		{
			SEMA_ERROR(expr, "An 'out' variable may not be read from.");
			return false;
		}
		asm_reg_add_input(block, arg);
	}
	if (is_write)
	{
		decl->var.is_written = true;
		if (decl->var.in_param)
		{
			SEMA_ERROR(expr, "An 'in' variable may not be written to.");
			return false;
		}
		asm_reg_add_output(block, arg);
	}
	Type *type = type_flatten(decl->type);
	if (type_is_pointer_type(type))
	{
		type = type_uptr->canonical;
	}
	if (type_is_integer(type))
	{
		if (!arg_type.ireg_bits)
		{
			if (arg_type.is_address)
			{
				SEMA_ERROR(expr, "You need to pass the variable by address.");
				return false;
			}
			SEMA_ERROR(expr, "An integer variable was not expected here.");
			return false;
		}
		if (!sema_reg_int_suported_type(arg_type, type))
		{
			unsigned bits = arg_bits_max(arg_type.ireg_bits, 0);
			ASSERT0(bits);
			SEMA_ERROR(expr, "%s is not supported in this position, convert it to a valid type, like %s.",
					   type_quoted_error_string(decl->type), type_quoted_error_string(type_int_signed_by_bitsize(bits)));
			return false;
		}
		return true;
	}
	if (type_is_float(type))
	{
		if (!arg_type.float_bits)
		{
			if (arg_type.is_address)
			{
				SEMA_ERROR(expr, "You need to pass the variable by address.");
				return false;
			}
			SEMA_ERROR(expr, "A floating point variable was not expected here.");
			return false;
		}
		if (!sema_reg_float_suported_type(arg_type, type))
		{
			SEMA_ERROR(expr, "%s is not supported in this position, convert it to a valid type.",
					   type_quoted_error_string(decl->type));
			return false;
		}
		return true;

	}
	SEMA_ERROR(expr, "%s is not supported as an argument.", type_quoted_error_string(decl->type));
	return false;
}

static inline bool sema_check_asm_memvar(SemaContext *context, AsmInlineBlock *block, AsmInstruction *instr, AsmArgType arg_type, Expr *expr)
{
	ExprAsmArg *arg = &expr->expr_asm_arg;
	const char *name = arg->ident.name;
	Decl *decl = sema_resolve_symbol(context, name, NULL, expr->span);
	if (!decl) return false;
	ASSERT0(arg->kind == ASM_ARG_MEMVAR);
	arg->ident.ident_decl = decl;
	if (decl->decl_kind != DECL_VAR)
	{
		RETURN_SEMA_ERROR(expr, "Expected a global or local variable.");
	}
	if (IS_OPTIONAL(decl))
	{
		RETURN_SEMA_ERROR(expr, "Optional variables are not allowed in asm.");
	}
	bool is_write = arg_type.is_write;
	bool is_read = !arg_type.is_write || arg_type.is_readwrite;
	arg->ident.is_input = !is_write;
	if (is_read)
	{
		decl->var.is_read = true;
		if (decl->var.out_param)
		{
			RETURN_SEMA_ERROR(expr, "An 'out' variable may not be read from.");
		}
		asm_reg_add_input(block, arg);
	}
	if (is_write)
	{
		decl->var.is_written = true;
		if (decl->var.in_param)
		{
			RETURN_SEMA_ERROR(expr, "An 'in' variable may not be written to.");
		}
		asm_reg_add_output(block, arg);
	}
	if (!arg_type.is_address)
	{
		RETURN_SEMA_ERROR(expr, "This slot does not accept an address.");
	}
	return true;
}

static inline bool sema_check_asm_arg_value(SemaContext *context, AsmInlineBlock *block, AsmInstruction *instr, AsmArgType arg_type, Expr *expr)
{
	Expr *inner = exprptr(expr->expr_asm_arg.expr_id);
	if (!sema_analyse_expr(context, inner)) return false;
	if (expr_is_const_int(inner)) return sema_check_asm_arg_const_int(context, block, instr, arg_type, expr, inner);
	if (arg_type.is_write)
	{
		SEMA_ERROR(expr, "This position is written to, you can't use an expression for that.");
		return false;
	}
	Type *type = type_flatten(inner->type);
	if (type_is_pointer_type(type)) type = type_uptr->canonical;
	if (type_is_integer(type))
	{
		if (!sema_reg_int_suported_type(arg_type, type))
		{
			SEMA_ERROR(expr, "%s is not valid for this slot.", type_quoted_error_string(inner->type));
			return false;
		}
		asm_reg_add_input(block, &expr->expr_asm_arg);
		expr->type = type;
		return true;
	}
	if (type_is_float(type))
	{
		if (!sema_reg_float_suported_type(arg_type, type))
		{
			SEMA_ERROR(expr, "%s is not valid for this slot.", type_quoted_error_string(inner->type));
			return false;
		}
		asm_reg_add_input(block, &expr->expr_asm_arg);
		expr->type = type;
		return true;
	}
	TODO
}
static inline bool sema_check_asm_arg(SemaContext *context, AsmInlineBlock *block, AsmInstruction *instr, AsmArgType arg_type, Expr *expr)
{
	switch (expr->expr_asm_arg.kind)
	{
		case ASM_ARG_INT:
			return true;
		case ASM_ARG_REG:
			return sema_check_asm_arg_reg(context, block, instr, arg_type, expr);
		case ASM_ARG_ADDR:
			return sema_check_asm_arg_addr(context, block, instr, arg_type, expr);
		case ASM_ARG_VALUE:
			return sema_check_asm_arg_value(context, block, instr, arg_type, expr);
		case ASM_ARG_REGVAR:
			return sema_check_asm_var(context, block, instr, arg_type, expr);
		case ASM_ARG_MEMVAR:
			return sema_check_asm_memvar(context, block, instr, arg_type, expr);
		case ASM_ARG_ADDROF:
			TODO
			break;
	}
	UNREACHABLE
}

bool sema_analyse_asm(SemaContext *context, AsmInlineBlock *block, Ast *asm_stmt)
{
	ASSERT0(compiler.platform.asm_initialized);

	AsmInstruction *instr = asm_instr_by_name(asm_stmt->asm_stmt.instruction);
	if (!instr) RETURN_SEMA_ERROR(asm_stmt, "Unknown instruction");

	// Check arguments
	Expr **args = asm_stmt->asm_stmt.args;
	unsigned expected_params = instr->param_count;
	unsigned arg_count = vec_size(args);
	if (expected_params != arg_count)
	{
		RETURN_SEMA_ERROR(asm_stmt, "Too %s arguments to instruction '%s', expected %d.",
						  expected_params > arg_count ? "few" : "many",
						  instr->name, expected_params);
	}

	// Sema check each argument.
	for (unsigned i = arg_count; i > 0; i--)
	{
		if (!sema_check_asm_arg(context, block, instr, instr->param[i - 1], args[i - 1])) return false;
	}

	// Add clobbers
	sema_add_clobbers(block, &instr->mask);
	return true;
}