Systems and methods for performing instructions specifying ternary tile logic operations
Disclosed embodiments relate to systems and methods for performing instructions specifying ternary tile operations. In one example, a processor includes fetch and decode circuitry to fetch and decode an instruction specifying a ternary tile operation, and locations of destination and first, second, and third source matrices, each of the matrices having M rows by N columns; and execution circuitry to respond to the decoded instruction by, for each equal-sized group of K elements of the specified first, second, and third source matrices, generate K results by performing the ternary tile operation in parallel on K corresponding elements of the specified first, second, and third source matrices, and store each of the K results to a corresponding element of the specified destination matrix, wherein corresponding elements of the specified source and destination matrices occupy a same relative position within their associated matrix.
1. A processor comprising:
a plurality of vector registers;
a matrix operations accelerator comprising, separate from the plurality of vector registers, a first two-dimensional tile register to store a first source matrix, a second two-dimensional tile register to store a second source matrix, and a third two-dimensional tile register to store a third source matrix;
decode circuitry to decode an instruction specifying a ternary tile operation, a location of a destination matrix, the first two-dimensional tile register to store the first source matrix, the second two-dimensional tile register to store the second source matrix, and the third two-dimensional tile register to store the third source matrix into a decoded instruction; and
execution circuitry to respond to the decoded instruction to, for each equal-sized group of K elements of the specified first, second, and third source matrices, cause the matrix operations accelerator to generate K results by performing the ternary tile operation on K corresponding elements of the specified first, second, and third source matrices, and store each of the K results to a corresponding element of the specified destination matrix, wherein corresponding elements of the specified source and destination matrices occupy a same relative position within their associated matrices.
2. The processor of claim 1 , wherein each of the K elements in the equal-sized groups of elements of the specified first, second and third source matrices has an element width of one bit, and wherein the ternary tile operation is a bitwise operation.
3. The processor of claim 1 , wherein each of the matrices have M rows by N columns and each of the equal-sized groups of elements of the specified first, second, and third source matrices is an O by P submatrix, wherein M is an integer multiple of O and N is an integer multiple of P.
4. The processor of claim 1 , wherein each of the equal-sized groups of K elements of the specified first, second, and third source matrices has a total width that equals a width of a parallel execution datapath of the execution circuitry.
5. The processor of claim 1 , wherein the first two-dimensional tile register to store the first source matrix is also the location of the destination matrix to which the K generated results are stored.
6. The processor of claim 1 , wherein the ternary tile operation is a logical operation.
7. The processor of claim 6 , wherein the instruction comprises an immediate that specifies the logical operation from a plurality of logical operations.
8. The processor of claim 1 , wherein the instruction comprises an immediate that specifies the ternary tile operation from a plurality of ternary tile operations.
9. A method comprising:
decoding, using decode circuitry of a processor comprising a plurality of vector registers and a matrix operations accelerator comprising, separate from the plurality of vector registers, a first two-dimensional tile register storing a first source matrix, a second two-dimensional tile register storing a second source matrix, and a third two-dimensional tile register storing a third source matrix, an instruction specifying a ternary tile operation, a location of a destination matrix, and the first two-dimensional tile register storing the first source matrix, the second two-dimensional tile register storing the second source matrix, and the third two-dimensional tile register storing the third source matrix into a decoded instruction; and
executing the decoded instruction with execution circuitry of the processor to, for each equal-sized group of K elements of the specified first, second, and third source matrices, cause the matrix operations accelerator to generate K results by performing the ternary tile operation in parallel on K corresponding elements of the specified first, second, and third source matrices, and store each of the K results to a corresponding element of the specified destination matrix, wherein corresponding elements of the specified source and destination matrices occupy a same relative position within their associated matrices.
10. The method of claim 9 , wherein each of the K elements in the equal-sized groups of elements of the specified first, second and third source matrices has an element width of one bit, and wherein the ternary tile operation is a bitwise operation.
11. The method of claim 9 , wherein each of the matrices have M rows by N columns and each of the equal-sized groups of elements of the specified first, second, and third source matrices is an O by P submatrix, wherein M is an integer multiple of O and N is an integer multiple of P.
12. The method of claim 9 , wherein each of the equal-sized groups of K elements of the specified first, second, and third source matrices has a total width that equals a width of a parallel execution datapath of the execution circuitry.
13. The method of claim 9 , wherein the first two-dimensional tile register to store the first source matrix is also the location of the destination matrix to which the K generated results are stored.
14. The method of claim 9 , wherein the ternary tile operation is a logical operation.
15. The method of claim 14 , wherein the instruction comprises an immediate that specifies the logical operation from a plurality of logical operations.
16. The method of claim 9 , wherein the instruction comprises an immediate that specifies the ternary tile operation from a plurality of ternary tile operations.
17. A system comprising:
a matrix operations accelerator comprising, separate from a plurality of vector registers, a first two-dimensional tile register to store a first source matrix, a second two-dimensional tile register to store a second source matrix, and a third two-dimensional tile register to store a third source matrix; and
a processor comprising:
the plurality of vector registers;
decode circuitry to decode an instruction specifying a ternary tile operation, a location of a destination matrix, the first two-dimensional tile register to store the first source matrix, the second two-dimensional tile register to store the second source matrix, and the third two-dimensional tile register to store the third source matrix into a decoded instruction, and
execution circuitry to respond to the decoded instruction to, for each equal-sized group of K elements of the specified first, second, and third source matrices, cause the matrix operations accelerator to generate K results by performing the ternary tile operation in parallel on K corresponding elements of the specified first, second, and third source matrices, and store each of the K results to a corresponding element of the specified destination matrix, wherein corresponding elements of the specified source and destination matrices occupy a same relative position within their associated matrices.
18. The system of claim 17 , wherein each of the K elements in the equal-sized groups of elements of the specified first, second and third source matrices has an element width of one bit, and wherein the ternary tile operation is a bitwise operation.
19. The system of claim 17 , wherein the ternary tile operation is a logical operation.
20. The system of claim 19 , wherein the instruction comprises an immediate that specifies the logical operation from a plurality of logical operations.
21. The system of claim 17 , wherein the instruction comprises an immediate that specifies the ternary tile operation from a plurality of ternary tile operations.
22. The system of claim 17 , wherein the first two-dimensional tile register to store the first source matrix is also the location of the destination matrix to which the K generated results are stored.