Apparatus for performing multiply-add operations on packed data

An apparatus for use in a computer system comprising:

• a storage area having stored therein a first packed data and a second packed data each containing initial data elements, each of said initial data elements in said first packed data having only one corresponding initial data element in said second packed data;
• a circuit, coupled to said storage area, operating in response to a signal, said circuit comprising:
• a first means for multiplying together said corresponding initial data elements in said first packed data and said second packed data to generate corresponding intermediate data elements, said intermediate data elements being divided into a number of pairs;
• a second means for generating a plurality of result data elements, a first of said plurality of result data elements representing the sum of said intermediate data elements in a first of said number of pairs, a second of said plurality of result data elements representing the sum of said intermediate result data elements in a second of said number of pairs;
• and
• a third means for storing said plurality of result data elements as a third packed data in said storage area, wherein each element in the third packed data is the result of a multiply-add operation on a separate set of four of said initial data elements.

A processor comprising:

• a register file to respectively store in separate registers a first and second packed data each containing initial data elements, each of said initial data elements in said first packed data corresponding to a different initial data element in said second packed data;
• a decoder to decode an instruction that specifies said first and second packed data as operands;
• and
• a circuit coupled to said register file and said decoder, said circuit in response to the instruction to,
• simultaneously multiply together only said corresponding initial data elements in said first and second packed data to generate corresponding intermediate data elements, said intermediate data elements being paired into a plurality of sets,
• arithmetically combine the intermediate data elements in each of said plurality of sets to generate a plurality of result data elements wherein each of said plurality of result data elements provides a higher precision than said initial data elements, and wherein a first and second of said plurality of result data elements respectively represent the addition of said intermediate data elements in a first and second of said plurality of sets, and
• store as a result of executing the single instruction said plurality of result data elements in one register of said register file as a third packed data, wherein each element in the third packed data is the result of a multiply-add operation on a separate set of four of said initial data elements.

A processor comprising:

• a register file to store a first packed data and a second packed data each containing initial data elements, each of said initial data elements in said first packed data corresponding to a different initial data element in said second packed data to create a plurality of pairs of initial data elements;
• a decoder to decode an instruction that specifies said first and said packed data as operands;
• and
• a first circuit coupled to said decoder and responsive to said instruction, said first circuit including,
• a plurality of multipliers, each of said plurality of multipliers coupled to receive a different one of said plurality of pairs of initial data elements, said plurality of multipliers being divided into a plurality of multiplier pairs,
• a plurality of arithmetic units, each of said plurality of arithmetic units coupled to a different one of said plurality of multiplier pairs, wherein two of said arithmetic units are adders, and
• a second circuit, coupled to said plurality of arithmetic units, to cause a third packed data to be stored in said register file for use as an operand by another instruction, each data element in said third packed data stores the result of a different one of said plurality of arithmetic units, and each data element in said third packed data provides a higher precision than said initial data elements.

A processor comprising:

• a register file to store a first packed data and a second packed data each containing four initial data elements, each of said initial data elements in said first packed data corresponding to a different initial data element in said second packed data to create four pairs of initial data elements,
• a decoder to decode an instruction that specifies said first and second packed data as operands, and
• a circuit coupled to said register file and said decoder, said circuit in response to the instruction to,
• simultaneously multiply together said corresponding initial data elements in said first and second packed data to generate corresponding intermediate data elements, said intermediate data elements being paired into two sets,
• add the intermediate data elements in each of said two sets to generate two result data elements, wherein each of said two result data elements provides a higher precision than said initial data elements, and
• store as a result of executing the single instruction said two result data elements in one register of said register file as a third packed data, wherein said third packed data contains only said two result data elements.

A processor comprising:

• a register file to store a first packed data and a second packed data each containing four initial data elements, each of said initial data elements in said first packed data corresponding to a different initial data element in said second packed data to create four pairs of initial data elements,
• a decoder to decode an instruction that specifies said first and said packed data as operands, and
• a first circuit coupled to said decoder and responsive to said decoded instruction, said first circuit including,
• a plurality of multipliers, each of said plurality of multipliers coupled to receive a different one of said four pairs of initial data elements, said plurality of multipliers being divided into a first and second multiplier pair,
• a first and second adder respectively coupled to said first and second multiplier pair, and
• a second circuit, coupled to said first and second adders, to cause a third packed data containing only a first and second result data elements to be stored in said register file for use as an operand by another instruction, each of said result data elements representing the result of a different one of said first and second adders, and each of said result data elements providing a higher precision than said initial data elements.

A processor comprising:

• a first and second storage areas to have a first and second packed data, respectively, each packed data having a first, second, third, and fourth data element;
• a multiply-add circuit coupled to said first and second storage areas, said multiply-add circuit having a first, second, third, and fourth multiplier, each said multiplier to receive a corresponding set of said data elements;
• said multiply-add circuit further including a first adder coupled to said first multiplier and second multiplier, and a second adder coupled to said third multiplier and fourth multiplier;
• and
• a third storage area coupled to said adders, said third storage area having a first and second field for saving outputs of said first and second adders, respectively, as first and second unaccumulated data elements of a third packed data.

A processor comprising:

• a first storage area operable to have stored therein a first packed data containing at least an A1, an A2, an A3, and an A4 element;
• a second storage area operable to have stored therein a second packed data containing at least a B1, a B2, a B3, and a B4 element;
• a multiply circuit including
• a first multiplier coupled to said first storage area to receive said A1 and coupled to said second storage area to receive said B1;
• a second multiplier coupled to said first storage area to receive said A2 and coupled to said second storage are to receive said B2;
• a third multiplier coupled to said first storage area to receive said A3 and coupled to said second storage area to receive said B3;
• a fourth multiplier coupled to said first storage area to receive said A4 and coupled to said second storage area to receive said B4;
• a first adder coupled to said first multiplier and said second multiplier;
• a second adder coupled to said third multiplier and said fourth multiplier;
• and
• a third storage area coupled to said first adder and said second adder, said third storage area having at least a first field and a second field, said first field for saving an unaccumulated output of said first adder as a first unaccumulated data element of a third packed data, said second field for saving an unaccumulated output of said second adder as a second unaccumulated data element of said third packed data.