Apparatuses and methods to accelerate vector multiplication of vector elements having matching indices
Methods and apparatuses relating to accelerating vector multiplication. In one embodiment, an apparatus includes a first buffer to store a first cache line of indices for elements of a first vector, a second buffer to store a second cache line of indices for elements of a second vector, a comparison unit to compare each index of the first cache line of indices with each index of the second cache line of indices, a plurality of multipliers to each multiply an element from the first vector and an element from the second vector for an index match from the comparison unit to produce a product, and an adder to add together the product from each of the plurality of multipliers.
1. An apparatus comprising:
a first buffer to store a first cache line of indices for elements of a first vector;
a second buffer to store a second cache line of indices for elements of a second vector;
a comparison unit to compare each index of the first cache line of indices with each index of the second cache line of indices;
a plurality of multipliers to each multiply an element from the first vector and an element from the second vector for an index match from the comparison unit to produce a product; and
an adder to add together the product from each of the plurality of multipliers.
2. The apparatus of claim 1 , wherein a first streamer is to provide an index and its element from a data storage device to the first buffer and a second streamer is to provide an index and its element from the data storage device to the second buffer.
3. The apparatus of claim 1 , wherein the indices of the first cache line and the second cache line are not in index order.
4. The apparatus of claim 1 , wherein the comparison unit is to compare each index of the first cache line of indices with each index of the second cache line of indices in a single clock cycle of a processor.
5. The apparatus of claim 1 , wherein the first cache line of indices for elements of the first vector also includes each index's element.
6. The apparatus of claim 1 , wherein the plurality of multipliers are a plurality of multiplier-accumulator units.
7. The apparatus of claim 1 , further comprising logic to notify a requesting processor core that operations on all elements of the first vector and the second vector are completed.
8. The apparatus of claim 1 , wherein the comparison unit is to return each index of the first cache line to the first buffer and each index of the second cache line to the second buffer for non-matching indices.
9. A method comprising:
retrieving a first cache line of indices for elements of a first vector stored in a first buffer;
retrieving a second cache line of indices for elements of a second vector stored in a second buffer;
comparing each index of the first cache line of indices with each index of the second cache line of indices with a comparison unit;
multiplying an element from the first vector and an element from the second vector for each of a plurality of multipliers for an index match from the comparison unit to produce a product; and
adding together the product from each of the plurality of multipliers with an adder.
10. The method of claim 9 , further comprising:
providing an index and its element from a data storage device to the first buffer with a first streamer, and
providing an index and its element from the data storage device to the second buffer with a second streamer.
11. The method of claim 9 , wherein the indices of the first cache line and the second cache line are not in index order.
12. The method of claim 9 , wherein the comparing is in a single clock cycle of a processor.
13. The method of claim 9 , wherein the first cache line of indices for elements of the first vector also includes each index's element.
14. The method of claim 9 , wherein the plurality of multipliers are a plurality of multiplier-accumulator units.
15. The method of claim 9 , further comprising notifying a requesting processor core that operations on all elements of the first vector and the second vector are completed.
16. The method of claim 9 , further comprising returning each index of the first cache line to the first buffer and each index of the second cache line to the second buffer for non-matching indices.
17. A system comprising:
a data storage device to store a first vector and a second vector;
a first buffer to store a first cache line of indices for elements of the first vector from the data storage device;
a second buffer to store a second cache line of indices for elements of the second vector from the data storage device;
a comparison unit to compare each index of the first cache line of indices with each index of the second cache line of indices;
a plurality of multipliers to each multiply an element from the first vector and an element from the second vector for an index match from the comparison unit to produce a product; and
an adder to add together the product from each of the plurality of multipliers.
18. The system of claim 17 , wherein a first streamer is to provide an index and its element from the data storage device to the first buffer and a second streamer is to provide an index and its element from the data storage device to the second buffer.
19. The system of claim 17 , wherein the indices of the first cache line and the second cache line are not in index order.
20. The system of claim 17 , wherein the comparison unit is to compare each index of the first cache line of indices with each index of the second cache line of indices in a single clock cycle of a processor.
21. The system of claim 17 , wherein the first cache line of indices for elements of the first vector also includes each index's element.
22. The system of claim 17 , wherein the plurality of multipliers are a plurality of multiplier-accumulator units.
23. The system of claim 17 , further comprising logic to notify a requesting processor core that operations on all elements of the first vector and the second vector are completed.
24. The system of claim 17 , wherein the comparison unit is to return each index of the first cache line to the first buffer and each index of the second cache line to the second buffer for non-matching indices.