Apparatus and method for vector horizontal add of signed/unsigned words and doublewords

A processor comprising:

• a decoder to decode an instruction to generate a decoded instruction, the instruction including an opcode and operands identifying a plurality of packed doublewords, wherein the opcode is to identify the plurality of packed doublewords as signed or unsigned packed doublewords and the operands include at most a source register operand and a destination register operand;
• a source register specified by the source register operand to store a first plurality of packed doublewords;
• execution circuitry to execute the decoded instruction, the execution circuitry comprising:
• operand selection circuitry to identify first and second packed doublewords from the source register in accordance with the operands and opcode of the instruction, wherein the first and second packed doublewords are selected from non-adjacent fixed locations in the source register;
• adder circuitry to add the first and second packed doublewords to generate a temporary sum;
• a temporary storage of at least 33 bits and coupled to the adder circuitry to store the temporary sum generated by the adder circuitry, the temporary storage storing at most a single instance of the temporary sum;
• and
• saturation circuitry to saturate the temporary sum if necessary to generate a final result;
• and
• a destination register specified by the destination register operand to store the final result as a packed result doubleword in a designated data element position, wherein the designated data element position corresponds to a data element position of one of the first and second packed doublewords, and any remaining bits in the destination register are set to zero,
• wherein respective operations of the operand selection circuitry, the adder circuitry, the temporary storage, and the saturation circuitry are all performed responsive to a single execution of the decoded instruction by the execution circuitry.

A method comprising:

• decoding an instruction by a decoder, the instruction including an opcode and operands identifying a plurality of packed words, wherein the opcode is to identify the plurality of packed words as signed or unsigned packed words and the operands include at most a source register operand and a destination register operand;
• storing a first plurality of packed words in a source register specified by the source register operand;
• executing, by execution circuitry, a single instance of the decoded instruction to responsively perform operations of:
• identifying, by operand selection circuitry, first and second packed words from the source register in accordance with the operands and opcode of the instruction, wherein the first and second packed words are selected from non-adjacent fixed locations in the source register;
• adding, by adder circuitry, the first and second packed words to generate a temporary sum;
• storing the temporary sum in a temporary storage of at least 17 bits, the temporary storage storing at most a single instance of the temporary sum;
• saturating, by saturation circuitry, the temporary sum if necessary to generate a final result;
• storing the final result as a packed result word in a designated data element position of a destination register specified by the destination register operand, the designated data element position to correspond to a data element position of one of the first and second packed words;
• and
• setting any remaining bits in the destination register to zero.

A non-transitory machine-readable medium having program code stored thereon which, when executed by a machine, causes the machine to perform operations of:

• decoding an instruction by a decoder, the instruction including an opcode and operands identifying a plurality of packed words, wherein the opcode is to identify the plurality of packed words as signed or unsigned packed words and the operands include at most a source register operand and a destination register operand;
• storing a first plurality of packed words in a source register specified by the source register operand;
• executing, by execution circuitry, a single instance of the decoded instruction to responsively perform operations of:
• identifying, by operand selection circuitry, first and second packed words from a source register in accordance with the operands and opcode of the instruction, wherein the first and second packed words are selected from non-adjacent fixed locations in the source register;
• adding, by adder circuitry, the first and second packed words to generate a temporary sum;
• storing the temporary sum in a temporary storage of at least 17 bits, the temporary storage storing at most a single instance of the temporary sum;
• saturating, by saturation circuitry, the temporary sum if necessary to generate a final result;
• storing the final result as a packed result word in a designated data element position of a destination register specified by the destination register operand, the designated data element position to correspond to a data element position of one of the first and second packed words;
• and
• setting any remaining bits in the destination register to zero.