updates to many ships

[fleet.git] / ships / Alu3.ship

diff --git a/ships/Alu3.ship b/ships/Alu3.ship
index b3f892a..025d7eb 100644 (file)
--- a/ships/Alu3.ship
+++ b/ships/Alu3.ship
@@ -9,13 +9,38 @@ data  out:  out1
   shortcut to: in1
 data  out:  out2
   shortcut to: in2
-data  out:  out3
-  shortcut to: in3
 data  out:  outBits
 
 == Constants ========================================================
 == TeX ==============================================================
 
+{\tt Alu3} is a three-input adder which produces a pair of outputs in
+carry-save form.  It has no opcode input.
+
+This ship also contains a private ``bit fifo'' similar to the {\tt
+BitFifo} ship, except that only the dequeueing (output) interface is
+exposed to the programmer.  Each addition operation performed causes
+the lowest bit of the {\it save} output to be enqueued into the bit
+fifo.  This can be used to produce a very efficient multiplier; see
+the test case for this ship for more details.
+
+\subsection*{Semantics}
+
+When a value is present at each of {\tt in1}, {\tt in2} and {\tt in3},
+these three values are consumed.  The {\it carry} result of carry-save
+addition is placed in {\tt out1}, and the {\it save} result of
+carry-save addition is placed in {\tt out2}.
+
+\subsection*{To Do}
+
+Is the second output supposed to be shifted?
+
+Provide a way to clear/flush the internal bitfifo.
+
+Do we even need this?  Can we do the same thing with {\tt Lut3} and
+{\tt BitFifo} together?
+
+
 == Fleeterpreter ====================================================
 boolean mode = false;
 BitFifo.BitStorage outBits = new BitFifo.BitStorage(74);
@@ -30,19 +55,16 @@ public void service() {
       box_in3.dataReadyForShip() &&
       outBits.hasSpace(1) &&
       box_out1.readyForDataFromShip() &&
-      box_out2.readyForDataFromShip() &&
-      box_out3.readyForDataFromShip()) {
+      box_out2.readyForDataFromShip()) {
       long v1 = box_in1.removeDataForShip();
       long v2 = box_in2.removeDataForShip();
       long v3 = box_in3.removeDataForShip();
       long o1, o2, o3;
       o1 = ((v1 & v2) | (v2 & v3) | (v1 & v3))/* << 1*/;
       o2 = (v1 ^ v2 ^ v3) >> 1;
-      o3 = 0;
       outBits.add((v1 ^ v2 ^ v3) & 0x1L, 1);
       box_out1.addDataFromShip(o1);
       box_out2.addDataFromShip(o2);
-      box_out3.addDataFromShip(o3);
   }
 }
 
@@ -59,7 +81,6 @@ public void service() {
   reg [(`DATAWIDTH-1):0] keep_in3;     initial keep_in3 = 0;
   reg                    have_out1;    initial have_out1 = 0;
   reg                    have_out2;    initial have_out2 = 0;
-  reg                    have_out3;    initial have_out3 = 0;
   reg [73:0] bitstorage; initial bitstorage = 0;
   reg [7:0] bitstorage_count;          initial bitstorage_count = 0;
   reg wrote;                           initial wrote = 0;
@@ -76,8 +97,6 @@ public void service() {
       `onwrite(out1_r, out1_a) have_out1 <= 0; end
     end else if (have_out2) begin
       `onwrite(out2_r, out2_a) have_out2 <= 0; end
-    end else if (have_out3) begin
-      `onwrite(out3_r, out3_a) have_out3 <= 0; end
     end else if (!have_in1) begin
       `onread(in1_r, in1_a) have_in1 <= 1; keep_in1 <= in1_d; end
     end else if (!have_in2) begin
@@ -89,12 +108,10 @@ public void service() {
           out2_d                       <= { 1'b0, (keep_in1[(`DATAWIDTH-1):1] ^
                                                    keep_in2[(`DATAWIDTH-1):1] ^
                                                    keep_in3[(`DATAWIDTH-1):1]) };
-          out3_d                       <= 0;
         bitstorage[bitstorage_count]  = (keep_in1[0] ^ keep_in2[0] ^ keep_in3[0]);
         bitstorage_count             <= bitstorage_count+1;
         have_out1 <= 1;
         have_out2 <= 1;
-        have_out3 <= 1;
         have_in1  <= 0;
         have_in2  <= 0;
         have_in3  <= 0;
@@ -162,7 +179,6 @@ alu3.in2:      [1] notify lut3.out; [*] take, deliver, notify lut3.out;
 alu3.in3:      literal 0; deliver; [*] take, deliver;
 alu3.out1:     [74] take, sendto alu3.in1;
 alu3.out2:     [74] take, sendto alu3.in3;
-alu3.out3:     [74] take;
 alu3.outBits:  [*] take, sendto debug.in;