pattern shiftadd
//
// Transforms add/sub+shift pairs that result from expressions such as data[s*W +C +:W2]
// specifically something like: out[W2-1:0] = data >> (s*W +C)
// will be transformed into: out[W2-1:0] = (data >> C) >> (s*W)
// this can then be optimized using peepopt_shiftmul_right.pmg
//

match shift
	select shift->type.in($shift, $shiftx, $shr)
	filter !port(shift, \B).empty()
endmatch

// the right shift amount
state <SigSpec> shift_amount
// log2 scale factor in interpreting of shift_amount
// due to zero padding on the shift cell's B port
state <int> log2scale
// zeros at the MSB position make it unsigned
state <bool> msb_zeros

code shift_amount log2scale msb_zeros
	shift_amount = port(shift, \B);

	log2scale = 0;
	while (shift_amount[0] == State::S0) {
		shift_amount.remove(0);
		if (shift_amount.empty()) reject;
		log2scale++;
	}

	msb_zeros = 0;
	while (shift_amount.bits().back() == State::S0) {
		msb_zeros = true;
		shift_amount.remove(GetSize(shift_amount) - 1);
		if (shift_amount.empty()) reject;
	}
endcode

state <bool>    var_signed
state <SigSpec> var_signal
// offset: signed constant value c in data[var+c +:W1] (constant shift-right amount)
state <int>     offset

match add
	// either data[var+c +:W1] or data[var-c +:W1]
	select add->type.in($add, $sub)
	index <SigSpec> port(add, \Y) === shift_amount

	// one must be constant, the other is variable
	choice <IdString> constport {\A, \B}
	select !port(add, constport).empty()
	select port(add, constport).is_fully_const()
	define <IdString> varport (constport == \A ? \B : \A)

	// only optimize for constants up to a fixed width. this prevents cases
	// with a blowup in internal term size and prevents larger constants being
	// casted to int incorrectly
	select (GetSize(port(add, constport)) <= 24)

	// if a value of var is able to wrap the output, the transformation might give wrong results
	// an addition/substraction can at most flip one more bit than the largest operand (the carry bit)
	// as long as the output can show this bit, no wrap should occur (assuming all signed-ness make sense)
	select ( GetSize(port(add, \Y)) > max(GetSize(port(add, \A)), GetSize(port(add, \B))) )

	define <bool> varport_A (varport == \A)
	define <bool> is_sub add->type.in($sub)

	define <bool> constport_signed param(add, !varport_A ? \A_SIGNED : \B_SIGNED).as_bool()
	define <bool> varport_signed param(add, varport_A ? \A_SIGNED : \B_SIGNED).as_bool();
	define <bool> const_negative (constport_signed && (port(add, constport).bits().back() == State::S1))
	define <bool> offset_negative ((is_sub && varport_A) ^ const_negative)

	// checking some value boundaries as well:
	// data[...-c +:W1] is fine for any signed var (pad at LSB, all data still accessible)
	// unsigned shift may underflow (eg var-3 with var<3) -> cannot be converted
	// data[...+c +:W1] is only fine for +var(add) and var unsigned
	// (+c cuts lower C bits, making them inaccessible, a signed var could try to access them)
	// either its an add or the variable port is A (it must be positive)
	select (add->type.in($add) || varport == \A)

	// -> data[var+c +:W1] (with var signed) is illegal
	filter !(!offset_negative && varport_signed)
	// -> data >> (var-c) (with var unsigned) is illegal
	filter !(offset_negative && !varport_signed)

	// state-variables are assigned at the end only:
	// shift the log2scale offset in-front of add to get true value: (var+c)<<N -> (var<<N)+(c<<N)
	set offset ( (port(add, constport).as_int(constport_signed) << log2scale) * ( (is_sub && varport_A) ? -1 : 1 ) )

	set var_signed varport_signed
	set var_signal add->getPort(varport)
endmatch

code
{
	// positive constant offset with a signed variable (index) cannot be handled
	// the above filter should get rid of this case but 'offset' is calculated differently
	// due to limitations of state-variables in pmgen
	// it should only differ if previous passes create invalid data
	log_assert(!(offset>0 && var_signed));

	SigSpec old_a = port(shift, \A); // data
	std::string location = shift->get_src_attribute();

	if(shiftadd_max_ratio>0 && offset<0 && -offset*shiftadd_max_ratio > old_a.size()) {
		log_warning("at %s: candiate for shiftadd optimization (shifting '%s' by '%s - %d' bits) " 
					"was ignored to avoid high resource usage, see help peepopt\n", 
					location.c_str(), log_signal(old_a), log_signal(var_signal), -offset);
		reject;
	}

	did_something = true;
	log("shiftadd pattern in %s: shift=%s, add/sub=%s, offset: %d\n", \
			log_id(module), log_id(shift), log_id(add), offset);

	SigSpec new_a;
	if(offset<0) {
		// data >> (...-c) transformed to {data, c'X} >> (...)
		SigSpec padding( (shift->type.in($shiftx) ? State::Sx : State::S0), -offset );
		new_a.append(padding);
		new_a.append(old_a);
	} else {
		// data >> (...+c) transformed to data[MAX:c] >> (...)
		if(offset < GetSize(old_a)) { // some signal bits left?
			new_a.append(old_a.extract_end(offset));
		} else {
			// warn user in case data is empty (no bits left)
			if (location.empty())
				location = shift->name.str();
			if(shift->type.in($shiftx))
				log_warning("at %s: result of indexed part-selection is always constant (selecting from '%s' with index '%s + %d')\n", \
							location.c_str(), log_signal(old_a), log_signal(var_signal), offset);
			else
				log_warning("at %s: result of shift operation is always constant (shifting '%s' by '%s + %d' bits)\n", \
							location.c_str(), log_signal(old_a), log_signal(var_signal), offset);
		}
	}

	SigSpec new_b = {var_signal, SigSpec(State::S0, log2scale)};
	if (msb_zeros || !var_signed)
		new_b.append(State::S0);

	shift->setPort(\A, new_a);
	shift->setParam(\A_WIDTH, GetSize(new_a));
	shift->setPort(\B, new_b);
	shift->setParam(\B_WIDTH, GetSize(new_b));
	blacklist(add);
	accept;
}
endcode