Intel Corporation

Tal Uliel

Elmoustapha Ould-Ahmed-Vall

Robert Valentine

Bret L Toll

Jesus Corbal

Zeev Sperber

Amit Gradstein

Mark Jay Charney

Thomas Willhalm

Mostafa Hagog

Boris Bolshem

A method is described that includes fetching an instruction. The method further includes decoding the instruction. The instruction specifies an operation, a first operand and a second operand. The method further includes fetching the first and second operands of the instruction. The first and second operands are each composed of a plurality of larger chunks having constituent elements. The method further includes performing the operation specified by the instruction including generating a resultant composed of a plurality of larger chunks having constituent elements. The generating of the resultant includes selecting for each element in the resultant a contiguous group of bits from a same positioned chunk of the first operand as the chunk of the element in the resultant, the contiguous group of bits being identified by a same positioned element of the second operand as the element in the resultant.

Functional unit for instruction execution pipeline capable of shifting different chunks of a packed data operand by different amounts

An apparatus and method are described for down-converting from a source operand to a destination operand with masking. For example, a method according to one embodiment includes the following operations: reading a source operand value to be down-converted from a first value to a down-converted value and stored in a destination location; reading each mask register bit stored in a mask register, the mask register bit(s) indicating whether to perform a masking operation or a conversion operation on the source operand value; if the mask register bit(s) indicates that a masking operation is to be performed, then performing a specified masking operation and storing the results of the masking operation in the destination location; and if the mask register bit indicates that a masking operation is not to be performed, then down-converting the source operand value and storing the down-converted value in the specified destination location.

Apparatus and method for down conversion of data types

Instructions and logic provide vector compress and rotate functionality. A processor may include a mask register, a decoder, and an execution unit. The mask register may include a data field, wherein the data field corresponds to an element location in a vector. The decoder may be coupled to the mask register. The decoder may decode an instruction to obtain a decoded instruction. The decoded instruction may specify a vector source, the mask register, a vector destination, and a vector destination offset location. The execution unit is coupled to the decoder. The execution unit may read an unmasked value in the data field; copy an vector element from the vector source to a location adjacent to the element; change the unmasked value to a masked value; determine that the vector destination is full; store a vector destination operand associated with the vector destination in a memory; and re-execute the instruction using the masked value and the vector destination offset location.

Instruction and logic to provide vector compress and rotate functionality

According to one embodiment, a code optimizer is configured to receive first code having a program loop implemented with scalar instructions to store values of a first array to a second array based on values of a third array and to generate second code representing the program loop using at least one vector instruction. The second code include a shuffle instruction to shuffle elements of the first array based on the third array using a shuffle table in a vector manner and a store instruction to store the shuffled elements of the first array in the second array.

Methods to optimize a program loop via vector instructions using a shuffle table

A processing device to provide vectorization of conditional loops includes vector physical registers to store a source vector having a first plurality of n data fields, and a destination vector comprising a second plurality of data fields corresponding to the first plurality of data fields, wherein each of the second plurality of data fields corresponds to a mask value in a vector conditions mask. The processing device includes a decode stage to decode a first processor instruction specifying a vector expand operation and a data partition size, and execution units to set elements of the source vector to n count values, obtain a decisions vector, generate the vector conditions mask according to the decisions vector, and copy data from consecutive vector elements in the source vector, into unmasked vector elements of the destination vector, without copying data from the source vector into masked vector elements of the destination vector.

Instructions and logic to vectorize conditional loops

Instructions and logic provide vector compress and rotate functionality. Some embodiments, responsive to an instruction specifying: a vector source, a mask, a vector destination and destination offset, read the mask, and copy corresponding unmasked vector elements from the vector source to adjacent sequential locations in the vector destination, starting at the vector destination offset location. In some embodiments, the unmasked vector elements from the vector source are copied to adjacent sequential element locations modulo the total number of element locations in the vector destination. In some alternative embodiments, copying stops whenever the vector destination is full, and upon copying an unmasked vector element from the vector source to an adjacent sequential element location in the vector destination, the value of a corresponding field in the mask is changed to a masked value. Alternative embodiments zero elements of the vector destination, in which no element from the vector source is copied.

An apparatus and method are described for shuffling data elements from source registers to a destination register. For example, a method according to one embodiment includes the following operations: reading each mask bit stored in a mask data structure, the mask data structure containing mask bits associated with data elements of a destination register, the values usable for determining whether a masking operation or a shuffle operation should be performed on data elements stored within a first source register and a second source register; for each data element of the destination register, if a mask bit associated with the data element indicates that a shuffle operation should be performed, then shuffling data elements from the first source register and the second source register to the specified data element within the destination register; and if the mask bit indicates that a masking operation should be performed, then performing a specified masking operation with respect to the data element of the destination register.

Apparatus and method for shuffling floating point or integer values

An apparatus and method are described for permuting data elements with masking. For example, a method according to one embodiment includes the following operations: reading values from a mask data structure to determine whether masking is implemented for each data element of a destination operand; if masking not implemented for a particular data element, then selecting data elements from a first source operand and a second source operand based on index values stored in destination operand to be copied to data element positions within the destination operand, wherein any one of the data elements from either the first source operand and the second source operand may be copied to any one of the data element positions within the destination operand; and if masking is implemented for a particular data element of the destination operand, then performing a designated masking operation with respect to that particular data element.

Apparatus and method for performing permute operations

Instructions and logic provide vectorization of conditional loops. A vector expand instruction has a parameter to specify a source vector, a parameter to specify a conditions mask register, and a destination parameter to specify a destination vector to hold n consecutive vector elements, each of the plurality of n consecutive vector elements having a same variable partition size of m bytes. In response to the processor instruction, data is copied from consecutive vector elements in the source vector, and expanded into unmasked vector elements of the specified destination vector, without copying data into masked vector elements of the destination vector, wherein n varies responsive to the processor instruction executed. The source vector may be a register and the destination vector may be in memory. Some embodiments store counts of the condition decisions. Alternative embodiments may store other data, for example such as target addresses, or table offsets, or indicators of processing directives, etc.

An apparatus and method are described for permuting data elements with masking. For example, a method according to one embodiment includes the following operations: reading values from a mask data structure to determine whether masking is implemented for each data element of a destination operand; if masking not implemented for a particular data element, then selecting data elements from the destination operand and a second source operand based on index values stored in a first source operand to be copied to data element positions within the destination operand, wherein any one of the data elements from either the destination operand and the second source operand may be copied to any one of the data element positions within the destination operand; if masking is implemented for a particular data element of the destination operand, then performing a designated masking operation with respect to that particular data element.

Apparatus and method for performing a permute operation

A processor includes a decoder to receive an instruction that indicates first and second source packed data operands and at least one shift count. An execution unit is operable, in response to the instruction, to store a result packed data operand. Each result data element includes a first least significant bit (LSB) portion of a first data element of a corresponding pair of data elements in a most significant bit (MSB) portion, and a second MSB portion of a second data element of the corresponding pair in a LSB portion. One of the first LSB portion of the first data element and the second MSB portion of the second data element has a corresponding shift count number of bits. The other has a number of bits equal to a size of a data element of the first source packed data minus the corresponding shift count.

Packed two source inter-element shift merge processors, methods, systems, and instructions

According to one embodiment, a code optimizer is configured to receive first code having a program loop implemented with scalar instructions to store values of a first array to a second array based on values of a third array. The code optimizer is configured to generate second code representing the program loop with vector instructions including a shuffle instruction and a store instruction, the store instruction to shuffle using a shuffle table elements of the first array based on the second array in a vector manner, the store instruction to store using a mask store table the shuffled elements in the third array in a vector manner.

Methods to optimize a program loop via vector instructions using a shuffle table and a mask store table

Methods, apparatus, instructions and logic provide SIMD vector sub-byte decompression functionality. Embodiments include shuffling a first and second byte into the least significant portion of a first vector element, and a third and fourth byte into the most significant portion. Processing continues shuffling a fifth and sixth byte into the least significant portion of a second vector element, and a seventh and eighth byte into the most significant portion. Then by shifting the first vector element by a first shift count and the second vector element by a second shift count, sub-byte elements are aligned to the least significant bits of their respective bytes. Processors then shuffle a byte from each of the shifted vector elements' least significant portions into byte positions of a destination vector element, and from each of the shifted vector elements' most significant portions into byte positions of another destination vector element.

Growing community of inventors

Tel Aviv, Israel

Tal Uliel