data in: in3
data out: out1
+ shortcut to: in1
data out: out2
+ shortcut to: in2
+data out: out3
+ shortcut to: in3
+data out: outBits
== Constants ========================================================
== TeX ==============================================================
-This ship performs addition of three inputs, producing two output
-values in carry-save form. To complete the addition, send the two
-output values to an Alu2 with opcode ADD. For summing a set of four
-or more numbers, Alu3 followed by Alu2 is often faster than repeated
-use of Alu2.
-
== Fleeterpreter ====================================================
+boolean mode = false;
+BitFifo.BitStorage outBits = new BitFifo.BitStorage(74);
public void service() {
if (box_in1.dataReadyForShip() &&
box_in2.dataReadyForShip() &&
box_in3.dataReadyForShip() &&
+ outBits.hasSpace(1) &&
box_out1.readyForDataFromShip() &&
- box_out2.readyForDataFromShip()) {
- long v1 = box_in1.removeDataForShip();
- long v2 = box_in2.removeDataForShip();
- long v3 = box_in3.removeDataForShip();
- long o1 = ((v1 & v2) | (v2 & v3) | (v1 & v3)) << 1;
- long o2 = v1 ^ v2 ^ v3;
+ box_out2.readyForDataFromShip() &&
+ box_out3.readyForDataFromShip()) {
+ long v1 = box_in1.removeDataForShip();
+ long v2 = box_in2.removeDataForShip();
+ long v3 = box_in3.removeDataForShip();
+ long o1, o2, o3;
+ mode = true;
+ if (!mode) {
+ o1 = v1;
+ o2 = v2 >>> 1;
+ o3 = ((v2 & 0x1L) == 0) ? 0 : v1;
+ } else {
+ 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);
+ mode = !mode;
+ }
+
+ if (box_outBits.readyForDataFromShip() &&
+ outBits.size() >= 37) {
+ box_outBits.addDataFromShip(outBits.get(37));
}
}
== FPGA ==============================================================
- reg have_a;
- reg [(`DATAWIDTH-1):0] a;
- reg have_b;
- reg [(`DATAWIDTH-1):0] b;
- reg have_c;
- reg [(`DATAWIDTH-1):0] c;
- reg have_out1;
- reg have_out2;
+ reg mode; initial mode = 0;
+ reg have_in1; initial have_in1 = 0;
+ reg have_in2; initial have_in2 = 0;
+ reg have_in3; initial have_in3 = 0;
+ reg [(`DATAWIDTH-1):0] keep_in1; initial keep_in1 = 0;
+ reg [(`DATAWIDTH-1):0] keep_in2; initial keep_in2 = 0;
+ 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;
+ reg [7:0] bitstorage_count; initial bitstorage_count = 0;
+ reg wrote; initial wrote = 0;
always @(posedge clk) begin
+ wrote = 0;
if (have_out1) begin
`onwrite(out1_r, out1_a) have_out1 <= 0; end
- end
- if (have_out2) begin
+ end else if (have_out2) begin
`onwrite(out2_r, out2_a) have_out2 <= 0; end
- end
-
- if (!have_out1 && !have_out2) begin
- if (!have_a) begin
- `onread(in1_r, in1_a) have_a <= 1; a <= in1_d; end
- end
- if (!have_b) begin
- `onread(in2_r, in2_a) have_b <= 1; b <= in2_d; 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
+ `onread(in2_r, in2_a) have_in2 <= 1; keep_in2 <= in2_d; end
+ end else if (!have_in3) begin
+ `onread(in3_r, in3_a) have_in3 <= 1; keep_in3 <= in3_d; end
+ end else begin
+ mode = 1;
+ if (mode == 0) begin
+ out1_d <= keep_in1;
+ out2_d <= { 1'b0, keep_in2[(`DATAWIDTH-1):1] };
+ out3_d <= (keep_in2[0]==0) ? 0 : keep_in1;
+ end else begin
+ out1_d <= { ((keep_in1 & keep_in2) | (keep_in2 & keep_in3) | (keep_in1 & keep_in3)) };
+ out2_d <= (keep_in1 ^ keep_in2 ^ keep_in3) >> 1;
+ out3_d <= 0;
+ bitstorage_count <= bitstorage_count+1;
+ bitstorage[bitstorage_count] = (keep_in1[0] ^ keep_in2[0] ^ keep_in3[0]);
+ outBits_d = bitstorage[(`DATAWIDTH-1):0];
+ wrote = 1;
end
- if (!have_c) begin
- `onread(in3_r, in3_a) have_c <= 1; c <= in3_d; end
- end
-
- if (have_a && have_b && have_c) begin
- out1_d <= ((a & b) | (b & c) | (a & c)) << 1;
- out2_d <= a ^ b ^ c;
have_out1 <= 1;
have_out2 <= 1;
- end
+ have_out3 <= 1;
+ have_in1 <= 0;
+ have_in2 <= 0;
+ have_in3 <= 0;
+// mode <= ~mode;
end
+
+ if (!wrote && bitstorage_count >= `DATAWIDTH) begin
+ `onwrite(outBits_r, outBits_a) begin
+ bitstorage_count <= bitstorage_count - `DATAWIDTH;
+ bitstorage = bitstorage >> `DATAWIDTH;
+ end
+ end
end
+end
+
+
+== Test ========================================================================
+#expect 100488372224
+#expect 8
+
+#ship alu3 : Alu3
+#ship debug : Debug
+1000000: sendto alu3.in1;
+1200000: sendto alu3.in2;
+0: sendto alu3.in3;
+0: sendto alu3.in1;
+0: sendto alu3.in2;
+alu3.in1: [*] take, deliver;
+alu3.in2: [*] take, deliver;
+alu3.in3: [*] take, deliver;
+alu3.out1: [74] take, sendto alu3.in1;
+ [74] take, sendto alu3.in1;
+alu3.out2: [74] take, sendto alu3.in2;
+ [74] take, sendto alu3.in2;
+alu3.out3: [74] take, sendto alu3.in3;
+ [74] take, sendto alu3.in3;
+alu3.outBits: [2] take, sendto debug.in;
+debug.in: [*] take, deliver;
== Contributors =========================================================
+Amir Kamil <kamil@cs.berkeley.edu>
Adam Megacz <megacz@cs.berkeley.edu>