摘要:
An apparatus and method for efficient matrix alignment in a systolic array. For example, one embodiment of a processor comprises: a first set of physical tile registers to store first matrix data in rows or columns; a second set of physical tile registers to store second matrix data in rows or columns; a decoder to decode a matrix instruction identifying a first input matrix, a first offset, a second input matrix, and a second offset; and execution circuitry, responsive to the matrix instruction, to read a subset of rows or columns from the first set of physical tile registers in accordance with the first offset, spanning multiple physical tile registers from the first set if indicated by the first offset to generate a first input matrix and the execution circuitry to read a subset of rows or columns from the second set of physical tile registers in accordance with the second offset, spanning multiple physical tile registers from the second set if indicated by the second offset to generate a second input matrix; and the execution circuitry to perform an arithmetic operation with the first and second input matrices in accordance with an opcode of the matrix instruction.
摘要:
Vector blend and permute functionality are provided, responsive to instructions specifying: a destination vector register comprising fields to store vector elements, a first vector register, a vector element size, a second vector register, and a third operand. Indices are read from fields in the second register. Each index has a first selector portion and a second selector portion. Corresponding unmasked vector elements are stored to fields of the destination register, wherein each vector element, responsive to the respective first selector portion having a first value, is copied to an intermediate vector from a corresponding data field of the first register, and responsive to the respective first selector portion having a second value, is copied to the intermediate vector from a corresponding data field of the third operand. Then unmasked data fields of the destination are replaced by data fields in the intermediate vector indexed by the corresponding second selector portions.
摘要:
Instructions and logic provide SIMD permute controls with leading zero count functionality. Some embodiments include processors with a register with a plurality of data fields, each of the data fields to store a second plurality of bits. A destination register has corresponding data fields, each of these data fields to store a count of the number of most significant contiguous bits set to zero for corresponding data fields. Responsive to decoding a vector leading zero count instruction, execution units count the number of most significant contiguous bits set to zero for each of data fields in the register, and store the counts in corresponding data fields of the first destination register. Vector leading zero count instructions can be used to generate permute controls and completion masks to be used along with the set of permute controls, to resolve dependencies in gather-modify-scatter SIMD operations.
摘要:
Systems, methods, and apparatuses for data speculation execution (DSX) are described. In some embodiments, a hardware apparatus for performing DSX comprises a hardware decoder to decode an instruction, the instruction to include an opcode and an operand to store a portion of a fallback address and an operand to store a stride value, execution hardware to execute the decoded instruction to initiate a data speculative execution (DSX) region by activating DSX tracking hardware to track speculative memory accesses and detect ordering violations in the DSX region, and storing the fallback address.
摘要:
Embodiments of systems, apparatuses, and methods for aggregate gather and scatter are disclosed. In some embodiments, a decoder to decode an instruction, wherein the instruction to include fields for an index of memory address locations, an immediate, and a starting destination register operand and identifier of additional destination registers; and execution circuitry to execute the decoded instruction to gather, from memory at locations indicated by the index of memory locations, data elements and stores them in multiple destination registers in sizes dictated by the immediate are described.
摘要:
A technique for decoding an instruction in a variable-length instruction set. In one embodiment, an instruction encoding is described, in which legacy, present, and future instruction set extensions are supported, and increased functionality is provided, without expanding the code size and, in some cases, reducing the code size.
摘要:
A technique to enable efficient instruction fusion within a computer system. In one embodiment, a processor logic delays the processing of a second instruction for a threshold amount of time if a first instruction within an instruction queue is fusible with the second instruction.
摘要:
A method of an aspect includes receiving a floating point round-off amount determination instruction. The instruction indicates a source of one or more floating point data elements, indicates a number of fraction bits after a radix point, and indicates a destination storage location. A result including one or more result floating point data elements is stored in the destination storage location in response to the floating point round-off amount determination instruction. Each of the one or more result floating point data elements includes a difference between a corresponding floating point data element of the source in a corresponding position, and a rounded version of the corresponding floating point data element of the source that has been rounded to the indicated number of the fraction bits. Other methods, apparatus, systems, and instructions are disclosed.
摘要:
A method is described that includes reading a first read mask from a first register. The method also includes reading a first vector operand from a second register or memory location. The method also includes applying the read mask against the first vector operand to produce a set of elements for operation. The method also includes performing an operation of the set elements. The method also includes creating an output vector by producing multiple instances of the operation's result. The method also includes reading a first write mask from a third register, the first write mask being different than the first read mask. The method also includes applying the write mask against the output vector to create a resultant vector. The method also includes writing the resultant vector to a destination register.
摘要:
A technique to enable efficient instruction fusion within a computer system. In one embodiment, a processor logic delays the processing of a second instruction for a threshold amount of time if a first instruction within an instruction queue is fusible with the second instruction.