VHDL behavioural subset.
This document describes the ALLIANCE VHDL subset for behavioural data flow descriptions.
In a data flow architecture only concurrent statements (except process) are supported. All sequential statements including loops, signal assignment, etc .. are to be banished.
Allowed concurrent statements are:
simple signal assignment
conditional signal assignment
selected signal assignment
concurrent assert statement
When using concurrent statements, an ordinary signal can be assigned only once. The value of the signal must be explicitly defined by the signal assignment (for example, in a selected signal assignment the value of the target signal is to be defined for every value that the select expression can take).
The above constraint may be felt as a hard restriction when designing distributed controlled hardware (precharged line, distributed multiplexer, etc ...). To hurdle this, VHDL uses a special feature: guarded-resolved signals.
A resolved signal is a signal declared with a resolved subtype (see vhdl(5)). A resolved subtype is a type combined with a resolution function. A resolved signal can be assigned by multiple signal assignments. Depending on the value of each driver, the resolution function determines the effective value of the signal.
A guarded signal is a resolved signal with drivers that can be disconected. A guarded signal must be assigned inside a block statement through a guarded signal assignment.
A distributed multiplexer may be described as :
signal Distributed_Mux : mux_bit bus; begin first_driver_of_mux : block (Sel1 = '1') begin Distributed_Mux <= guarded Data1; end block; second_driver_of_mux : block (Sel2 = '1') begin Distributed_Mux <= guarded Data2; end block;
LATCHES and REGISTERS
Sequential elements must be explicitly declared using the type reg_bit or reg_vector (and must be of kind register). A sequential element must be assigned inside a block statement by a guarded signal assignment.
Rising edge triggered D flip flop :
signal Reg : reg_bit register; begin flip_flop : block (ck = '1' and not ck'STABLE) begin Reg <= guarded Din; end block;
Level sensitive latch:
signal Reg : reg_bit register; begin latch : block (ck = '1') begin Lat <= guarded Din; end block;
In both cases, the guard expression must depend only on one signal if the description is to be processed by the logic synthetizer (boom + boog).
The following operators are only supported: not, and, or, xor, nor, nand, &, =, /=
They can be applied on all types supported by the subset. Other standard VHDL operators (+, -, >, <, ...) have not been implemented in the present release.
Timing information can be specified in behavioural descriptions using after clauses. However, those delays are currently only used for simulation. After clauses are supported but not used for synthesis and formal proof.
After clauses in block statements (for guarded signal assignments) are not supported for sequential elements (signals of kind register), but supported for bus elements (signals of kind bus). This is because the VHDL default disconnection time is null and this can generate unexpected behavior for sequential elements.
In selected signal assignment, only uniform delays are supported (the same After clause in all assignments).
Transport option is not supported. All delays are inertial delays.
Only two severity levels are supported in concurrent assert statements:
print a warning message if the assert condition is not satisfied.
print an error message if the assert condition is not satisfied. Then, stop the simulation.
Assert statements are ignored by the logic synthesis tool.
A special feature has been introduced in order to allow "don't care" specification when the logic synthtizer is targeted ( Beware : this feature is incompatible with the IEEE VHDL standard !!).
An output can be assigned to the value 'D' (don't care). This is taken into account by the logic synthesis tool in the optimization process. When the value of an output is 'D' the logic synthesis tool may turn it into a '1' or a '0'.
A 'D' value is understood as a '0' by the logic simulator (asimut).
Arraies other than bit_vector, reg_vector, mux_vector and wor_vector are not supported.
Here is the description of an adder with an accumulator register.
entity add_accu is port ( clk : in bit; command : in bit; data_in : in bit_vector (31 downto 0); data_out : out bit_vector (31 downto 0); cry_out : out bit; vdd : in bit; vss : in bit ); end add_accu; architecture data_flow of add_accu is signal eff_data : bit_vector (31 downto 0); -- effective operande signal adder_out : bit_vector (31 downto 0); -- adder's result signal adder_cry : bit_vector (32 downto 0); -- adder's carry signal accum_reg : reg_vector (31 downto 0) register; -- accumulator constant initialize : bit := '0'; constant accumulate : bit := '1'; begin -- select the effective operand with command select eff_data <= X"0000_0000" when initialize, accum_reg when accumulate; -- compute the result out of the adder adder_out <= eff_data xor data_in xor adder_cry; adder_cry (0) <= '0'; adder_cry (32 downto 1) <= (eff_data and adder_cry (31 downto 0)) or (data_in and adder_cry (31 downto 0)) or (aff_data and data_in ) ; -- write the result into the register on the rising edge of clk write : block (clk = '1' and not clk'STABLE) begin accum_reg <= guarded adder_out; end block; -- assign outputs cry_out <= adder_cry (32); data_out <= accum_reg ; -- check power supply assert (vdd = '1' and vss = '0') report "power sypply is missing" severity ERROR; end;
vhdl(5), vst(5), boom(1), loon(1), boog(1), asimut(1), proof(1)
boog.1alc(1), boom.1alc(1), ctl.5alc(5), fsm.5alc(5), loon.1alc(1), moka.1alc(1), proof.1alc(1), sxlib.5alc(5), syf.1alc(1), vasy.1alc(1), vasy.5alc(5), vhdl.5alc(5), vst.5alc(5), xsch.1alc(1).