Method and apparatus for vector processing
A processor includes a first vector processing unit including a first register file and first vector arithmetic logic unit; a second vector processing unit including a second register file and second vector arithmetic logic unit wherein the first register file has a first plurality of cross connections to the second vector arithmetic logic unit; wherein the second register file as a second plurality of cross connections to the first vector arithmetic logic unit.
1. An apparatus ( 302 ) for performing dynamically scalable single instruction multiple data (SIMD) operations comprising:
a first vector arithmetic logic unit ( 402 ) operably coupled to a first data vector (VRA) from a first input data bus, a second data vector (VRB) from a second vector data bus, and a third data vector (VRC) from a third vector data bus to produce a first vector result (VRD) wherein the first vector functional arithmetic logic unit ( 402 ) performs individual conditional operations on fields of VRA and VRB to produce VRD, and wherein the conditional operations are based on VRC;
a second vector arithmetic logic unit ( 401 ) operably coupled to the first data vector (VRA) from the first input data bus, the second data vector (VRB) from the second input data bus, and to the first vector arithmetic logic unit, wherein the individual conditional vector operations in the first vector arithmetic logic unit are conditionally controlled by a vector result from the second vector arithmetic logic unit.
2. The apparatus of claim 1 wherein the first vector arithmetic logic unit performs addition and subtraction.
3. The apparatus of claim 2 wherein the second vector arithmetic logic unit performs subtraction.
4. The apparatus of claim 1 , wherein a sign of the vector result from the second vector arithmetic logic unit controls individual conditional SIMD field operations of the fist vector arithmetic unit within a single clock cycle.
5. The apparatus of claim 1 , wherein the vector result from the second vector arithmetic logic unit are fed back as a vector input to the first vector arithmetic logic unit within a single clock cycle.
6. The apparatus of claim 1 , wherein the vector result from the second vector arithmetic logic unit is output as a second vector result (VRE).
7. An apparatus ( 302 ) for performing dynamically scalable SIMD operations, the apparatus ( 302 ) comprising:
a first vector arithmetic unit ( 402 ) having a first, second, and third data vector as inputs, the first vector arithmetic unit capable of performing at least vector addition and subtraction operations, wherein individual SIMD field operations in the first vector arithmetic unit are conditionally controlled by at least a portion of the third data vector input; and
a second vector arithmetic unit ( 401 ) operably coupled to the first vector arithmetic unit.
8. The apparatus of claim 7 wherein the first vector arithmetic unit performs at least vector addition and subtraction.
9. The apparatus of claim 7 , wherein packed bitwise values of the third data vector alter individual SIMD field operations of the first vector arithmetic unit within a single clock cycle.
10. The apparatus of claim 7 , wherein a dynamically scalable vector addition and subtraction operation is performed by the first and the second vector arithmetic units and output simultaneously, within a same clock cycle.
11. The apparatus of claim 7 , wherein the second vector arithmetic unit is a dedicated vector subtraction unit.
12. The apparatus of claim 7 , wherein a dynamically scalable vector conditional negate and add operation is performed within a single clock cycle by the first vector arithmetic unit.