FIELD OF INVENTION

The field of invention relates generally to computer processor architecture, and, more specifically, to matrix manipulation.

BACKGROUND

Matrices are increasingly important in many computing tasks such as machine learning and other bulk data processing.

BRIEF DESCRIPTION OF THE DRAWINGS

The present invention is illustrated by way of example and not limitation in the figures of the accompanying drawings, in which like references indicate similar elements and in which:

FIG. 1 illustrates an embodiment of configured tiles;

FIG. 2 illustrates several examples of matrix storage;

FIG. 3 illustrates an embodiment of a system utilizing a matrix (tile) operations accelerator;

FIGS. 4 and 5 show different embodiments of how memory is shared using a matrix operations accelerator;

FIG. 6 illustrates an embodiment of matrix multiply accumulate operation using tiles (“TMMA”);

FIG. 7 illustrates an embodiment of a subset of the execution of an iteration of a chained fused multiply accumulate instruction;

FIG. 8 illustrates an embodiment of a subset of the execution of an iteration of a chained fused multiply accumulate instruction;

FIG. 9 illustrates an embodiment of a subset of the execution of an iteration of a chained fused multiply accumulate instruction;

FIG. 10 illustrates an embodiment of a subset of the execution of an iteration of chained fused multiply accumulate instruction;

FIG. 11 illustrates power-of-two sized SIMD implementations wherein the accumulators use input sizes that are larger than the inputs to the multipliers according to an embodiment;

FIG. 12 illustrates an embodiment of a system utilizing matrix operations circuitry;

FIG. 13 illustrates an embodiment of a processor core pipeline supporting matrix operations using tiles;

FIG. 14 illustrates an embodiment of a processor core pipeline supporting matrix operations using tiles;

FIG. 15 illustrates an example of a matrix expressed in row major format and column major format;

FIG. 16 illustrates an example of usage of matrices (tiles);

FIG. 17 illustrates an embodiment a method of usage of matrices (tiles);

FIG. 18 illustrates an exemplary execution of a TILECONFIG instruction;

FIGS. 19(A)-(D) illustrate examples of register(s);

FIG. 20 illustrates an embodiment of a description of the matrices (tiles) to be supported;

FIG. 21 illustrates an embodiment of method performed by a processor to process a TILECONFIG instruction;

FIG. 22 illustrates a more detailed description of an execution of a TILECONFIG instruction using memory addressing;

FIG. 23 illustrates exemplary pseudocode for an execution of a TILECONFIG instruction;

FIG. 24 illustrates an exemplary execution of a TILEADD instruction;

FIG. 25 illustrates an embodiment of method performed by a processor to process a TILEADD instruction;

FIG. 26 illustrates an example process describing a method performed by a processor to process a TILEADD instruction;

FIG. 27 illustrates an example method for performing a TILEADD operation when the source matrix (tile) operands contain single-precision elements;

FIG. 28 illustrates an exemplary execution of a TILESUB instruction;

FIG. 29 illustrates an embodiment of method performed by a processor to process a TILESUB instruction;

FIG. 30 illustrates an example process describing a method performed by a processor to process a TILESUB instruction;

FIG. 31 illustrates an example method for performing a TILESUB operation when the source matrix (tile) operands contain single-precision elements;

FIG. 32 illustrates an exemplary execution of a TILEMUL instruction;

FIG. 33 illustrates an embodiment of method performed by a processor to process a TILEMUL instruction;

FIG. 34 illustrates an example process describing a method performed by a processor to process a TILEMUL instruction;

FIG. 35 illustrates an example method for performing a TILEMUL operation when the source matrix (tile) operands contain single-precision elements;

FIGS. 36A-C illustrate an exemplary instruction format;

FIG. 37 is a block diagram of a register architecture according to one embodiment of the invention;

FIGS. 38A-B illustrate the in-order pipeline and in-order core;

FIGS. 39A-B illustrate a block diagram of a more specific exemplary in-order core architecture, which core would be one of several logic blocks (including other cores of the same type and/or different types) in a chip;

FIG. 40 is a block diagram of a processor 4000 that may have more than one core, may have an integrated memory controller, and may have integrated graphics according to embodiments of the invention;

FIGS. 41-44 are block diagrams of exemplary computer architectures; and

FIG. 45 is a block diagram contrasting the use of a software instruction converter to convert binary instructions in a source instruction set to binary instructions in a target instruction set according to embodiments of the invention.

DETAILED DESCRIPTION

In the following description, numerous specific details are set forth. However, it is understood that embodiments of the invention may be practiced without these specific details. In other instances, well-known circuits, structures and techniques have not been shown in detail in order not to obscure the understanding of this description.

References in the specification to “one embodiment,” “an embodiment,” “an example embodiment,” etc., indicate that the embodiment described may include a particular feature, structure, or characteristic, but every embodiment may not necessarily include the particular feature, structure, or characteristic. Moreover, such phrases are not necessarily referring to the same embodiment. Further, when a particular feature, structure, or characteristic is described in connection with an embodiment, it is submitted that it is within the knowledge of one skilled in the art to affect such feature, structure, or characteristic in connection with other embodiments whether or not explicitly described.

In many mainstream processors, handling matrices is a difficult and/or instruction intensive task. For example, rows of a matrix could be put into a plurality of packed data (e.g., SIMD or vector) registers and then operated on individually. For example, an add two 8×2 matrices may require a load or gather into four packed data registers depending upon data sizes. Then a first add of packed data registers corresponding to a first row from each matrix is performed and a second add of packed data registers corresponding to a second row from each matrix is performed. Then the resulting packed data registers are scattered back to memory. While for small matrices this scenario may be acceptable, it is often not acceptable with larger matrices.

I. High-Level Discussion

Described herein are mechanisms to support matrix operations in computer hardware such as central processing units (CPUs), graphic processing units (GPUs), and accelerators. The matrix operations utilize 2-dimensional (2-D) data structures representing one or more packed regions of memory such as registers. Throughout this description, these 2-D data structures are referred to as tiles. Note that a matrix may be smaller than a tile (use less than all of a tile), or utilize a plurality of tiles (the matrix is larger than the size of any one tile). Throughout the description, matrix (tile) language is used to indicate operations performed using tiles that impact a matrix whether or not that matrix is larger than any one tile is not typically relevant.

Each tile may be acted upon by different operations such as those that are detailed herein and include, but are not limited to: matrix (tile) multiplication, tile add, tile subtract, tile diagonal, tile zero, tile transpose, tile dot product, tile broadcast, tile row broadcast, tile column broadcast, tile multiplication, tile multiplication and accumulation, tile move, etc. Additionally, support for operators such as the use of a scale and/or bias may be used with these operations or in support of non-numeric applications in the future, for instance, OpenCL “local memory,” data compression/decompression, etc.

Portions of storage (such as memory (non-volatile and volatile), registers, cache, etc.) are arranged into tiles of different horizontal and vertical dimensions. For example, a tile may have horizontal dimension of 4 (e.g., four rows of a matrix) and a vertical dimension of 8 (e.g., 8 columns of the matrix). Typically, the horizontal dimension is related element sizes (e.g., 2-, 4-, 8-, 16-, 32-, 64-, 128-bit, etc.). Multiple datatypes (single precision floating point, double precision floating point, integer, etc.) may be supported.

A. Exemplary Usage of Configured Tiles

FIG. 1 illustrates an embodiment of configured tiles. As shown, there are four tiles 111, 113, 115, and 117 that are loaded from application memory 101. In this example, tiles T0 111 and T1 113 have M rows and N columns with 4 element bytes (e.g., single precision data). Tiles T2 115 and T3 117 have M rows and N/2 columns with 8 element bytes (e.g., double precision data). As the double precision operands are twice the width of single precision, this configuration is consistent with a palette, used to provide tile options, supplying at least 4 names with total storage of 16*N*M bytes. Depending upon the instruction encoding scheme used, the number of tiles available varies.

In some embodiments, tile parameters are definable. For example, a “palette” is used to provide tile options. Exemplary options include, but are not limited to: the number of tile names, the number of bytes in a row of storage, the number of rows and columns in a tile, etc. For example, a maximum “height” (number of rows) of a tile may be defined as:

Tile Max Rows=Architected Storage/(The Number of Palette Names*The Number of Bytes per row)

As such, an application can be written such that a fixed usage of names will be able to take advantage of different storage sizes across implementations.

Configuration of tiles is done using a tile configuration (“TILECONFIG”) instruction, where a particular tile usage is defined in a selected palette. This declaration includes the number of tile names to be used, the requested number of rows and columns per name (tile), and, in some embodiments, the requested datatype of each tile. In some embodiments, consistency checks are performed during the execution of a TILECONFIG instruction to determine that it matches the restrictions of the palette entry.

B. Exemplary Tile Storage Types

FIG. 2 illustrates several examples of matrix storage. In (A), a tile is stored in memory. As shown, each “row” consists of four packed data elements. To get to the next “row,” a stride value is used. Note that rows may be consecutively stored in memory. Strided memory accesses allows for access of one row to then next when the tile storage does not map the underlying memory array row width.

Tile loads from memory and stores to memory are typically strided accesses from the application memory to packed rows of data. Exemplary TILELOAD and TILESTORE instructions, or other instruction references to application memory as a TILE operand in load-op instructions, are, in some embodiments, restartable to handle (up to) 2*rows of page faults, unmasked floating point exceptions, and/or interrupts per instruction.

In (B), a matrix is stored in a tile comprised of a plurality of registers such as packed data registers (single instruction, multiple data (SIMD) or vector registers). In this example, the tile is overlaid on three physical registers. Typically, consecutive registers are used, however, this need not be the case.

In (C), a matrix is stored in a tile in non-register storage accessible to a fused multiple accumulate (FMA) circuit used in tile operations. This storage may be inside of a FMA, or adjacent to it. Additionally, in some embodiments, discussed below, the storage may be for a data element and not an entire row or tile.

The supported parameters for the TMMA architecture are reported via CPUID. In some embodiments, the list of information includes a maximum height and a maximum SIMD dimension. Configuring the TMMA architecture requires specifying the dimensions for each tile, the element size for each tile and the palette identifier. This configuration is done by executing the TILECONFIG instruction.

Successful execution of a TILECONFIG instruction enables subsequent TILE operators. A TILERELEASEALL instruction clears the tile configuration and disables the TILE operations (until the next TILECONFIG instructions executes). In some embodiments, XSAVE, XSTORE, etc. are used in context switching using tiles. In some embodiments, 2 XCRO bits are used in XSAVE, one for TILECONFIF metadata and one bit corresponding to actual tile payload data.

TILECONFIG not only configures the tile usage, but also sets a state variable indicating that the program is in a region of code with tiles configured. An implementation may enumerate restrictions on other instructions that can be used with a tile region such as no usage of an existing register set, etc.

Exiting a tile region is typically done with the TILERELEASEALL instruction. It takes no parameters and swiftly invalidates all tiles (indicating that the data no longer needs any saving or restoring) and clears the internal state corresponding to being in a tile region.

In some embodiments, tile operations will zero any rows and any columns beyond the dimensions specified by the tile configuration. For example, tile operations will zero the data beyond the configured number of columns (factoring in the size of the elements) as each row is written. For example, with 64 byte rows and a tile configured with 10 rows and 12 columns, an operation writing FP32 elements would write each of the first 10 rows with 12*4 bytes with output/result data and zero the remaining 4*4 bytes in each row. Tile operations also fully zero any rows after the first 10 configured rows. When using 1K tile with 64 byte rows, there would be 16 rows, so in this example, the last 6 rows would also be zeroed.

In some embodiments, a context restore (e.g., XRSTOR), when loading data, enforces that the data beyond the configured rows for a tile will be maintained as zero. If there is no valid configuration, all rows are zeroed. XRSTOR of tile data can load garbage in the columns beyond those configured. It should not be possible for XRSTOR to clear beyond the number of columns configured because there is not an element width associated with the tile configuration.

Context save (e.g., XSAVE) exposes the entire TILE storage area when writing it to memory. If XRSTOR loaded garbage data in to the rightmost part of a tile, that data will be saved by XSAVE. XSAVE will write zeros for rows beyond the number specified for each tile.

In some embodiments, tile instructions are restartable. The operations that access memory allow restart after page faults. The computational instructions that deal with floating point operations also allow for unmasked floating point exceptions, with the masking of the exceptions controlled by a control and/or status register.

To support restarting instructions after these events, the instructions store information in the start registers detailed below.

II. Matrix (Tile) Operation Systems

A. Exemplary Hardware Support

FIG. 3 illustrates an embodiment of a system utilizing a matrix (tile) operations accelerator. In this illustration, a host processor/processing system 301 communicates commands 311 (e.g., matrix manipulation operations such as arithmetic or matrix manipulation operations, or load and store operations) to a matrix operations accelerator 307. However, this is shown this way for discussion purposes only. As detailed later, this accelerator 307 may be a part of a processing core. Typically, commands 311 that are tile manipulation operator instructions will refer to tiles as register-register (“reg-reg”) or register-memory (“reg-mem”) format. Other commands such as TILESTORE, TILELOAD, TILECONFIG, etc., do not perform data operations on a tile. Commands may be decoded instructions (e.g., micro-ops) or macro-instructions for the accelerator 307 to handle.

In this example, a coherent memory interface 303 is coupled to the host processor/processing system 301 and matrix operations accelerator 405 such that they can share memory. FIGS. 4 and 5 show different embodiments of how memory is shared using a matrix operations accelerator. As shown in FIG. 4, the host processor 401 and matrix operations accelerator circuitry 405 share the same memory 403. FIG. 5 illustrates an embodiment where the host processor 501 and matrix operations accelerator 505 do not share memory, but can access each other's memory. For example, processor 501 can access tile memory 507 and utilize its host memory 503 as normal. Similarly, the matrix operations accelerator 505 can access host memory 503, but more typically uses its own memory 507. Note these memories may be of different types.

The matrix operations accelerator 307 includes a plurality of FMAs 309 coupled to data buffers 305 (in some implementations, one or more of these buffers 305 are stored in the FMAs of the grid as shown). The data buffers 305 buffer tiles loaded from memory and/or tiles to be stored to memory (e.g., using a tileload or tilestore instruction). Data buffers may be, for example, a plurality of registers. Typically, these FMAs are arranged as a grid of chained FMAs 309 which are able to read and write tiles. In this example, the matrix operations accelerator 307 is to perform a matrix multiply operation using tiles T0, T1, and T2. At least one of tiles is housed in the FMA grid 309. In some embodiments, all tiles in an operation are stored in the FMA grid 309. In other embodiments, only a subset are stored in the FMA grid 309. As shown, T1 is housed and T0 and T2 are not. Note that A, B, and C refer to the matrices of these tiles which may or may not take up the entire space of the tile.

FIG. 6 illustrates an embodiment of matrix multiply accumulate operation using tiles (“TMMA”).

The number of rows in the matrix (TILE A 601) matches the number of serial (chained) FMAs comprising the computation's latency. An implementation is free to recirculate on a grid of smaller height, but the computation remains the same.

The source/destination vector comes from a tile of N rows (TILE C 605) and the grid of FMAs 611 performs N vector-matrix operations resulting in a complete instruction performing a matrix multiplication of tiles. Tile B 603 is the other vector source and supplies “broadcast” terms to the FMAs in each stage.

In operation, in some embodiments, the elements of matrix B (stored in a tile B 603) are spread across the rectangular grid of FMAs. Matrix B (stored in tile A 601) has its elements of a row transposed to match up with the columnar dimension of the rectangular grid of FMAs. At each FMA in the grid, an element of A and B are multiplied and added to the incoming summand (from above in the Figure) and the outgoing sum is passed to the next row of FMAs (or the final output).

The latency of a single step is proportional to K (row height of matrix B) and dependent TMMAs typically have enough source-destination rows (either in a single tile or across tile) to hide that latency. An implementation may also split the SIMD (packed data element) dimension M (row height of matrix A) across time steps, but this simply changes the constant that K is multiplied by. When a program specifies a smaller K than the maximum enumerated by the TMACC, an implementation is free to implement this with “masking” or “early outs.”

The latency of an entire TMMA is proportional to N*K. The repeat rate is proportional to N. The number of MACs per TMMA instruction is N*K*M.

FIG. 7 illustrates an embodiment of a subset of the execution of an iteration of a chained fused multiply accumulate instruction. In particular, this illustrates execution circuitry of an iteration of one packed data element position of the destination. In this embodiment, the chained fused multiply accumulate is operating on signed sources wherein the accumulator is 2× the input data size.

A first signed source (source 1 701) and a second signed source (source 2 703) each have four packed data elements. Each of these packed data elements stores signed data such as floating point data. A third signed source (source 3 709) has two packed data elements, each of which stores signed data. The sizes of the first and second signed sources 701 and 703 are half that of the third signed source (initial value or previous result) 709. For example, the first and second signed sources 701 and 703 could have 32-bit packed data elements (e.g., single precision floating point) while the third signed source 709 could have 64-bit packed data elements (e.g., double precision floating point).

In this illustration, only the two most significant packed data element positions of the first and second signed sources 701 and 703 and the most significant packed data element position of the third signed source 709 are shown. Of course, the other packed data element positions would also be processed.

As illustrated, packed data elements are processed in pairs. For example, the data of the most significant packed data element positions of the first and second signed sources 701 and 703 are multiplied using a multiplier circuit 705, and the data from second most significant packed data element positions of the first and second signed sources 701 and 703 are multiplied using a multiplier circuit 707. In some embodiments, these multiplier circuits 705 and 707 are reused for other packed data elements positions. In other embodiments, additional multiplier circuits are used so that the packed data elements are processed in parallel. In some contexts, parallel execution is done using lanes that are the size of the signed third source 709. The results of each of the multiplications are added using addition circuitry 711.

The result of the addition of the results of the multiplications is added to the data from most significant packed data element position of the signed source 3 709 (using a different adder 713 or the same adder 711).

Finally, the result of the second addition is either stored into the signed destination 715 in a packed data element position that corresponds to the packed data element position used from the signed third source 709, or passed on to the next iteration, if there is one. In some embodiments, a writemask is applied to this storage such that if a corresponding writemask (bit) is set, the storage happens, and, if not set, the storage does not happen.

FIG. 8 illustrates an embodiment of a subset of the execution of an iteration of a chained fused multiply accumulate instruction. In particular, this illustrates execution circuitry of an iteration of one packed data element position of the destination. In this embodiment, the chained fused multiply accumulate is operating on signed sources wherein the accumulator is 2× the input data size.

A first signed source (source 1 801) and a second signed source (source 2 803) each have four packed data elements. Each of these packed data elements stores signed data such as integer data. A third signed source (source 3 809) has two packed data elements, each of which stores signed data. The sizes of the first and second signed sources 801 and 803 are half that of the third signed source 809. For example, the first and second signed sources 801 and 803 could have 32-bit packed data elements (e.g., single precision floating point) the third signed source 809 could have 64-bit packed data elements (e.g., double precision floating point).

In this illustration, only the two most significant packed data element positions of the first and second signed sources 801 and 803 and the most significant packed data element position of the third signed source 809 are shown. Of course, the other packed data element positions would also be processed.

As illustrated, packed data elements are processed in pairs. For example, the data of the most significant packed data element positions of the first and second signed sources 801 and 803 are multiplied using a multiplier circuit 805, and the data from second most significant packed data element positions of the first and second signed sources 801 and 803 are multiplied using a multiplier circuit 807. In some embodiments, these multiplier circuits 805 and 807 are reused for other packed data elements positions. In other embodiments, additional multiplier circuits are used so that the packed data elements are processed in parallel. In some contexts, parallel execution is done using lanes that are the size of the signed third source (initial value or previous iteration result) 809. The results of each of the multiplications are added to the signed third source 809 using addition/saturation circuitry 811.

Addition/saturation (accumulator) circuitry 811 preserves a sign of an operand when the addition results in a value that is too big. In particular, saturation evaluation occurs on the infinite precision result between the multi-way-add and the write to the destination or next iteration. When the accumulator 811 is floating point and the input terms are integer, the sum of products and the floating point accumulator input value are turned into infinite precision values (fixed point numbers of hundreds of bits), the addition of the multiplication results and the third input is performed, and a single rounding to the actual accumulator type is performed.

Unsigned saturation means the output values are limited to a maximum unsigned number for that element width (all is). Signed saturation means a value is limited to the be in the range between a minimum negative number and a max positive number for that element width (for bytes for example, the range is from −128 (=−2{circumflex over ( )}7) to 127(=2{circumflex over ( )}7−1)).

The result of the addition and saturation check is stored into the signed result 815 in a packed data element position that corresponds to the packed data element position used from the signed third source 809, or passed on to the next iteration if there is one. In some embodiments, a writemask is applied to this storage such that if a corresponding writemask (bit) is set, the storage happens, and, if not set, the storage does not happen.

FIG. 9 illustrates an embodiment of a subset of the execution of an iteration of a chained fused multiply accumulate instruction. In particular, this illustrates execution circuitry of an iteration of one packed data element position of the destination. In this embodiment, the chained fused multiply accumulate is operating on a signed source and an unsigned source wherein the accumulator is 4× the input data size.

A first signed source (source 1 901) and a second unsigned source (source 2 903) each have four packed data elements. Each of these packed data elements has data such as floating point or integer data. A third signed source (initial value or result 915) has a packed data element of which stores signed data. The sizes of the first and second sources 901 and 903 are a quarter of the third signed source 915. For example, the first and second sources 901 and 903 could have 16-bit packed data elements (e.g., word) and the third signed source 915 could have 64-bit packed data elements (e.g., double precision floating point or 64-bit integer).

In this illustration, the four most significant packed data element positions of the first and second sources 901 and 903 and the most significant packed data element position of the third signed source 915 are shown. Of course, other packed data element positions would also be processed if there are any.

As illustrated, packed data elements are processed in quadruplets. For example, the data of the most significant packed data element positions of the first and second sources 901 and 903 are multiplied using a multiplier circuit 907, data from second most significant packed data element positions of the first and second sources 901 and 903 are multiplied using a multiplier circuit 907, data from third most significant packed data element positions of the first and second sources 901 and 903 are multiplied using a multiplier circuit 909, and data from the least significant packed data element positions of the first and second sources 901 and 903 are multiplied using a multiplier circuit 911. In some embodiments, the signed packed data elements of the first source 901 are sign extended and the unsigned packed data elements of the second source 903 are zero extended prior to the multiplications.

In some embodiments, these multiplier circuits 905-911 are reused for other packed data elements positions. In other embodiments, additional multiplier circuits are used so that the packed data elements are processed in parallel. In some contexts, parallel execution is done using lanes that are the size of the signed third source 915. The results of each of the multiplications are added using addition circuitry 911.

The result of the addition of the results of the multiplications is added to the data from most significant packed data element position of the signed source 3 915 (using a different adder 913 or the same adder 911).

Finally, the result 919 of the second addition is either stored into the signed destination in a packed data element position that corresponds to the packed data element position used from the signed third source 915, or passed to the next iteration. In some embodiments, a writemask is applied to this storage such that if a corresponding writemask (bit) is set, the storage happens, and, if not set, the storage does not happen.

FIG. 10 illustrates an embodiment of a subset of the execution of an iteration of chained fused multiply accumulate instruction. In particular, this illustrates execution circuitry of an iteration of one packed data element position of the destination. In this embodiment, the chained fused multiply accumulate is operating on a signed source and an unsigned source wherein the accumulator is 4× the input data size.

A first signed source (source 1 1001) and a second unsigned source (source 2 1003) each have four packed data elements. Each of these packed data elements stores data such as floating point or integer data. A third signed source (initial or previous result 1015) has a packed data element of which stores signed data. The sizes of the first and second sources 1001 and 1003 are a quarter of the third signed source 1015. For example, the first and second sources 1001 and 1003 could have 16-bit packed data elements (e.g., word) and the third signed source 1015 could have 64-bit packed data elements (e.g., double precision floating point or 64-bit integer).

In this illustration, the four most significant packed data element positions of the first and second sources 1001 and 1003 and the most significant packed data element position of the third signed source 1015 are shown. Of course, other packed data element positions would also be processed if there are any.

As illustrated, packed data elements are processed in quadruplets. For example, the data of the most significant packed data element positions of the first and second sources 1001 and 1003 are multiplied using a multiplier circuit 1007, data from second most significant packed data element positions of the first and second sources 1001 and 1003 are multiplied using a multiplier circuit 1007, data from third most significant packed data element positions of the first and second sources 1001 and 1003 are multiplied using a multiplier circuit 1009, and data from the least significant packed data element positions of the first and second sources 1001 and 1003 are multiplied using a multiplier circuit 1011. In some embodiments, the signed packed data elements of the first source 1001 are sign extended and the unsigned packed data elements of the second source 1003 are zero extended prior to the multiplications.

In some embodiments, these multiplier circuits 1005-1011 are reused for other packed data elements positions. In other embodiments, additional multiplier circuits are used so that the packed data elements are processed in parallel. In some contexts, parallel execution is done using lanes that are the size of the signed third source 1015. The result of the addition of the results of the multiplications is added to the data from most significant packed data element position of the signed source 3 1015 using addition/saturation circuitry 1013.

Addition/saturation (accumulator) circuitry 1013 preserves a sign of an operand when the addition results in a value that is too big or too small for signed saturation. In particular, saturation evaluation occurs on the infinite precision result between the multi-way-add and the write to the destination. When the accumulator 1013 is floating point and the input terms are integer, the sum of products and the floating point accumulator input value are turned into infinite precision values (fixed point numbers of hundreds of bits), the addition of the multiplication results and the third input is performed, and a single rounding to the actual accumulator type is performed.

The result 1019 of the addition and saturation check is stored into the signed destination in a packed data element position that corresponds to the packed data element position used from the signed third source 1015, or passed to the next iteration. In some embodiments, a writemask is applied to this storage such that if a corresponding writemask (bit) is set, the storage happens, and, if not set, the storage does not happen.

FIG. 11 illustrates power-of-two sized SIMD implementations wherein the accumulators use input sizes that are larger than the inputs to the multipliers according to an embodiment. Note the source (to the multipliers) and accumulator values may be signed or unsigned values. For an accumulator having 2× input sizes (in other words, the accumulator input value is twice the size of the packed data element sizes of the sources), table 1101 illustrates different configurations. For byte sized sources, the accumulator uses word or half-precision floating-point (HPFP) values that are 16-bit in size. For word sized sources, the accumulator uses 32-bit integer or single-precision floating-point (SPFP) values that are 32-bit in size. For SPFP or 32-bit integer sized sources, the accumulator uses 64-intenger or double-precision floating-point (DPFP) values that are 64-bit in size.

For an accumulator having 4× input sizes (in other words, the accumulator input value is four times the size of the packed data element sizes of the sources), table 1103 illustrates different configurations. For byte sized sources, the accumulator uses 32-bit integer or single-precision floating-point (SPFP) values that are 32-bit in size. For word sized sources, the accumulator uses 64-bit integer or double-precision floating-point (DPFP) values that are 64-bit in size in some embodiments.

For an accumulator having 8× input sizes (in other words, the accumulator input value is eight times the size of the packed data element sizes of the sources), table 1105 illustrates a configuration. For byte sized sources, the accumulator uses 64-bit integer.

As hinted at earlier, matrix operations circuitry may be included in a core, or as an external accelerator. FIG. 12 illustrates an embodiment of a system utilizing matrix operations circuitry. In this illustration, a plurality of entities are coupled with a ring interconnect 1245.

A plurality of cores 1201, 1203, 1205, and 1207 provide non-tile based instruction support. In some embodiments, matrix operations circuitry is provided in a core 1203, and in other embodiments matrix operations circuitry 1211 and 1213 is accessible on the ring interconnect 1245.

Additionally, one or more memory controllers 1223-1225 are provided to communicate with memory 1233 and 1231 on behalf of the cores and/or matrix operations circuitry.

FIG. 13 illustrates an embodiment of a processor core pipeline supporting matrix operations using tiles. Branch prediction and decode circuitry 1303 performs branch predicting of instructions, decoding of instructions, and/or both from instructions stored in instruction storage 1301. For example, instructions detailed herein may be stored in instruction storage. In some implementations, separate circuitry is used for branch prediction and in some embodiments, at least some instructions are decoded into one or more micro-operations, micro-code entry points, microinstructions, other instructions, or other control signals using microcode 1305. The branch prediction and decode circuitry 1303 may be implemented using various different mechanisms. Examples of suitable mechanisms include, but are not limited to, look-up tables, hardware implementations, programmable logic arrays (PLAs), microcode read only memories (ROMs), etc.

The branch prediction and decode circuitry 1303 is coupled to a rename/allocator circuitry 1307 which is coupled, in some embodiments, to scheduler circuitry 1309. In some embodiments, these circuits provide register renaming, register allocation, and/or scheduling functionality by performing one or more of: 1) renaming logical operand values to physical operand values (e.g., a register alias table in some embodiments), 2) allocating status bits and flags to the decoded instruction, and 3) scheduling the decoded instruction for execution on execution circuitry out of an instruction pool (e.g., using a reservation station in some embodiments).

The scheduler circuitry 1309 represents any number of different schedulers, including reservations stations, central instruction window, etc. The scheduler unit(s) scheduler circuitry 1309 is coupled to, or includes, physical register file(s) 1315. Each of the physical register file(s) 1315 represents one or more physical register files, different ones of which store one or more different data types, such as scalar integer, scalar floating point, packed integer, packed floating point, vector integer, vector floating point, status (e.g., an instruction pointer that is the address of the next instruction to be executed), tiles, etc. In one embodiment, the physical register file(s) 1315 comprises vector registers circuitry, write mask registers circuitry, and scalar registers circuitry. These register circuits may provide architectural vector registers, vector mask registers, and general purpose registers. The physical register file(s) 1315 is overlapped by a retirement circuit 1317 to illustrate various ways in which register renaming and out-of-order execution may be implemented (e.g., using a reorder buffer(s) and a retirement register file(s); using a future file(s), a history buffer(s), and a retirement register file(s); using a register maps and a pool of registers; etc.). The retirement circuit 1317 and the physical register file(s) 1315 are coupled to the execution circuit(s) 1311.

While register renaming is described in the context of out-of-order execution, it should be understood that register renaming may be used in an in-order architecture. While the illustrated embodiment of the processor may also include separate instruction and data cache units and a shared L2 cache unit, alternative embodiments may have a single internal cache for both instructions and data, such as, for example, a Level 1 (L1) internal cache, or multiple levels of internal cache. In some embodiments, the system may include a combination of an internal cache and an external cache that is external to the core and/or the processor. Alternatively, all of the cache may be external to the core and/or the processor.

The execution circuitry 1311 a set of one or more execution circuits 1321, 1323, and 1327 and a set of one or more memory access circuits 1325. The execution circuits 1321, 1323, and 1327 perform various operations (e.g., shifts, addition, subtraction, multiplication) and on various types of data (e.g., scalar floating point, packed integer, packed floating point, vector integer, vector floating point). While some embodiments may include a number of execution units dedicated to specific functions or sets of functions, other embodiments may include only one execution unit or multiple execution units that all perform all functions. The scalar circuitry 1321 performs scalar operations, the vector/SIMD circuitry 1323 performs vector/SIMD operations, and matrix operations circuitry 1327 performs matrix (tile) operations detailed herein.

The set of memory access units 1364 is coupled to the memory unit 1370, which includes a data TLB unit 1372 coupled to a data cache unit 1374 coupled to a level 2 (L2) cache unit 1376. In one exemplary embodiment, the memory access units 1364 may include a load unit, a store address unit, and a store data unit, each of which is coupled to the data TLB unit 1372 in the memory unit 1370. The instruction cache unit 1334 is further coupled to a level 2 (L2) cache unit 1376 in the memory unit 1370. The L2 cache unit 1376 is coupled to one or more other levels of cache and eventually to a main memory.

By way of example, the exemplary register renaming, out-of-order issue/execution core architecture may implement a pipeline as follows: 1) an instruction fetch circuit performs fetch and length decoding stages; 2) the branch and decode circuitry 1303 performs a decode stage; 3) the rename/allocator circuitry 1307 performs an allocation stage and renaming stage; 4) the scheduler circuitry 1309 performs a schedule stage; 5) physical register file(s) (coupled to, or included in, the scheduler circuitry 1307 and rename/allocate circuitry 1307 and a memory unit perform a register read/memory read stage; the execution circuitry 1311 performs an execute stage; 6) a memory unit and the physical register file(s) unit(s) perform a write back/memory write stage; 7) various units may be involved in the exception handling stage; and 8) a retirement unit and the physical register file(s) unit(s) perform a commit stage.

The core may support one or more instructions sets (e.g., the x86 instruction set (with some extensions that have been added with newer versions); the MIPS instruction set of MIPS Technologies of Sunnyvale, CA; the ARM instruction set (with optional additional extensions such as NEON) of ARM Holdings of Sunnyvale, CA), including the instruction(s) described herein. In one embodiment, the core 1390 includes logic to support a packed data instruction set extension (e.g., AVX1, AVX2), thereby allowing the operations used by many multimedia applications to be performed using packed data.

It should be understood that the core may support multithreading (executing two or more parallel sets of operations or threads), and may do so in a variety of ways including time sliced multithreading, simultaneous multithreading (where a single physical core provides a logical core for each of the threads that physical core is simultaneously multithreading), or a combination thereof (e.g., time sliced fetching and decoding and simultaneous multithreading thereafter such as in the Intel® Hyperthreading technology).

FIG. 14 illustrates an embodiment of a processor core pipeline supporting matrix operations using tiles. Branch prediction and decode circuitry 1403 performs branch predicting of instructions, decoding of instructions, and/or both from instructions stored in instruction storage 1401. For example, instructions detailed herein may be stored in instruction storage. In some implementations, separate circuitry is used for branch prediction and in some embodiments, at least some instructions are decoded into one or more micro-operations, micro-code entry points, microinstructions, other instructions, or other control signals using microcode 1405. The branch prediction and decode circuitry 1403 may be implemented using various different mechanisms. Examples of suitable mechanisms include, but are not limited to, look-up tables, hardware implementations, programmable logic arrays (PLAs), microcode read only memories (ROMs), etc.

The branch prediction and decode circuitry 1403 is coupled to a rename/allocator circuitry 1407 which is coupled, in some embodiments, to scheduler circuitry 1409. In some embodiments, these circuits provide register renaming, register allocation, and/or scheduling functionality by performing one or more of: 1) renaming logical operand values to physical operand values (e.g., a register alias table in some embodiments), 2) allocating status bits and flags to the decoded instruction, and 3) scheduling the decoded instruction for execution on execution circuitry out of an instruction pool (e.g., using a reservation station in some embodiments).

The scheduler circuitry 1409 represents any number of different schedulers, including reservations stations, central instruction window, etc. The scheduler unit(s) scheduler circuitry 1409 is coupled to, or includes, physical register file(s) 1415. Each of the physical register file(s) 1415 represents one or more physical register files, different ones of which store one or more different data types, such as scalar integer, scalar floating point, packed integer, packed floating point, vector integer, vector floating point, status (e.g., an instruction pointer that is the address of the next instruction to be executed), tiles, etc. In one embodiment, the physical register file(s) 1415 comprises vector registers circuitry, write mask registers circuitry, and scalar registers circuitry. These register circuits may provide architectural vector registers, vector mask registers, and general purpose registers. The physical register file(s) 1415 is overlapped by a retirement circuit 1417 to illustrate various ways in which register renaming and out-of-order execution may be implemented (e.g., using a reorder buffer(s) and a retirement register file(s); using a future file(s), a history buffer(s), and a retirement register file(s); using a register maps and a pool of registers; etc.). The retirement circuit 1417 and the physical register file(s) 1415 are coupled to the execution circuit(s) 1411.

While register renaming is described in the context of out-of-order execution, it should be understood that register renaming may be used in an in-order architecture. While the illustrated embodiment of the processor may also include separate instruction and data cache units and a shared L2 cache unit, alternative embodiments may have a single internal cache for both instructions and data, such as, for example, a Level 1 (L1) internal cache, or multiple levels of internal cache. In some embodiments, the system may include a combination of an internal cache and an external cache that is external to the core and/or the processor. Alternatively, all of the cache may be external to the core and/or the processor.

The execution circuitry 1411 a set of one or more execution circuits 1427 and a set of one or more memory access circuits 1425. The execution circuits 1427 perform matrix (tile) operations detailed herein.

The set of memory access units 1464 is coupled to the memory unit 1470, which includes a data TLB unit 1472 coupled to a data cache unit 1474 coupled to a level 2 (L2) cache unit 1476. In one exemplary embodiment, the memory access units 1464 may include a load unit, a store address unit, and a store data unit, each of which is coupled to the data TLB unit 1472 in the memory unit 1470. The instruction cache unit 1434 is further coupled to a level 2 (L2) cache unit 1476 in the memory unit 1470. The L2 cache unit 1476 is coupled to one or more other levels of cache and eventually to a main memory.

By way of example, the exemplary register renaming, out-of-order issue/execution core architecture may implement a pipeline as follows: 1) an instruction fetch circuit performs fetch and length decoding stages; 2) the branch and decode circuitry 1403 performs a decode stage; 3) the rename/allocator circuitry 1407 performs an allocation stage and renaming stage; 4) the scheduler circuitry 1409 performs a schedule stage; 5) physical register file(s) (coupled to, or included in, the scheduler circuitry 1407 and rename/allocate circuitry 1407 and a memory unit perform a register read/memory read stage; the execution circuitry 1411 performs an execute stage; 6) a memory unit and the physical register file(s) unit(s) perform a write back/memory write stage; 7) various units may be involved in the exception handling stage; and 8) a retirement unit and the physical register file(s) unit(s) perform a commit stage.

The core may support one or more instructions sets (e.g., the x86 instruction set (with some extensions that have been added with newer versions); the MIPS instruction set of MIPS Technologies of Sunnyvale, CA; the ARM instruction set (with optional additional extensions such as NEON) of ARM Holdings of Sunnyvale, CA), including the instruction(s) described herein. In one embodiment, the core 1490 includes logic to support a packed data instruction set extension (e.g., AVX1, AVX2), thereby allowing the operations used by many multimedia applications to be performed using packed data.

It should be understood that the core may support multithreading (executing two or more parallel sets of operations or threads), and may do so in a variety of ways including time sliced multithreading, simultaneous multithreading (where a single physical core provides a logical core for each of the threads that physical core is simultaneously multithreading), or a combination thereof (e.g., time sliced fetching and decoding and simultaneous multithreading thereafter such as in the Intel® Hyperthreading technology).

B. Layout

Throughout this description, data is expressed using row major data layout. Column major users should translate the terms according to their orientation. FIG. 15 illustrates an example of a matrix expressed in row major format and column major format. As shown, matrix A is a 2×3 matrix. When this matrix is stored in row major format, the data elements of a row are consecutive. When this matrix is stored in column major format, the data elements of a column are consecutive. It is a well-known property of matrices that A^T*B^T=(BA)^T, where superscript T means transpose. Reading column major data as row major data results in the matrix looking like the transpose matrix.

In some embodiments, row-major semantics are utilized in hardware, and column major data is to swap the operand order with the result being transposes of matrix, but for subsequent column-major reads from memory it is the correct, non-transposed matrix.

For example, if there are two column-major matrices to multiply:

a b
g i k
ag + bh ai + bj ak + bl

c d *
h j l =
cg + dh ci + dj ck + dl

e f

eg + fh ei + fj ek + fl

(3 × 2)
(2 × 3)
(3 × 3)

The input matrices would be stored in linear memory (column-major) as:

a c e
and
g h

b d f

ij

k l

kl.

Reading those matrices as row-major with dimensions 2×3 and 3×2, they would appear as:

a c e
and
g h

b d f
i j

k l

Swapping the order and matrix multiplying:

g h

a c e

ag + bh cg + dh eg + fh

i j
*

b d f =
ai + bj ci + dj ei + fj

k l

ak + bl ck + dl ek + fl

- the transpose matrix is out and can then be stored in in row-major order:

ag + bh
ai + bj
ak + bl

cg + dh
ci + dj
ck + dl

eg + fh
ei + fj
ek + fl

- and used in subsequent column major computations, it is the correct un-transposed matrix:

ag + bh
ai + bj
ak + bl

cg + dh
ci + dj
ck + dl

eg + fh
ei + fj
ek + fl

III. Exemplary Usage

FIG. 16 illustrates an example of usage of matrices (tiles). In this example, matrix C 1601 includes two tiles, matrix A 1603 includes one tile, and matrix B 1605 includes two tiles. This figure shows an example of the inner loop of an algorithm to compute a matrix multiplication. In this example, two result tiles, tmm0 and tmm1, from matrix C 1601 are used to accumulate the intermediate results. One tile from the A matrix 1603 (tmm2) is re-used twice as it multiplied by two tiles from the B matrix 1605. Pointers to load a new A tile and two new B tiles from the directions indicated by the arrows. An outer loop, not shown, adjusts the pointers for the C tiles.

The exemplary code as shown includes the usage of a tile configuration instruction and is executed to configure tile usage, load tiles, a loop to process the tiles, store tiles to memory, and release tile usage.

FIG. 17 illustrates an embodiment of usage of matrices (tiles). At 1701, tile usage is configured. For example, a TILECONFIG instruction is executed to configure tile usage including setting a numbers of rows and columns per tile. Typically, at least one matrix (tile) is loaded from memory at 1703.

IV. Exemplary Instructions

A. Tile Configuration

As discussed above, tile usage typically needs to be configured prior to use. For example, full usage of all rows and columns may not be needed. Not only does not configuring these rows and columns save power in some embodiments, but the configuration may be used to determine if an operation will generate an error. For example, a matrix multiplication of the form (N×M)*(L*N) will typically not work if M and L are not the same.

Detailed herein are embodiments of a matrix (tile) configuration (“TILECONFIG”) instruction and its execution. Prior to using matrices using tiles, in some embodiments, tile support is to be configured. For example, how many rows and columns per tile, tiles that are to be used, etc. are configured. A TILECONFIG instruction is an improvement to a computer itself as it provides for support to configure the computer to use a matrix accelerator (either as a part of a processor core, or as an external device). In particular, an execution of the TILECONFIG instruction causes a configuration to be retrieved from memory and applied to matrix (tile) settings within a matrix accelerator.

i. Exemplary Execution

FIG. 18 illustrates an exemplary execution of a TILECONFIG instruction. The TILECONFIG instruction format includes fields for an opcode and a memory address.

As illustrated, the TILECONFIG instruction uses the address as a pointer to a memory 1801 location containing the description of the matrices (tiles) to be supported 1803.

Execution circuitry 1811 of a processor/core 1805 performs the TILECONFIG by retrieving the description 1803 from memory 1801 via a memory controller 1815, configuring tiles for a palette (setting the number of rows and columns) in a tile configuration 1817, and marking that matrix support is in use. In particular, instruction execution resources 1811 are configured to use tiles as specified by setting tile configurations 1817. The instruction execution resources may also include a machine specific register or configuration register to indicate tile usage.

Tile configurations 1817 are set to indicate parameters per tile as indicated by the tile description 1803 via the execution of the TILECONFIG instruction. The set parameters are the number of rows and columns per tile. Additional values such as in-use and start values are also set. The tile configurations 1817 utilize one or more registers 1819 to store tile usage and configuration information.

ii. Exemplary tile storage

FIGS. 19(A)-(D) illustrate examples of register(s) 1819. FIG. 19(A) illustrates a plurality of registers 1819. As shown each tile (TMM0 1901 . . . TMMN 1903) has a separate register with each register storing a row and column size for that particular tile. StartK and StartM are stored in separate registers 1911 and 1913. One or more status registers 1915 are set (e.g., TILES_CONFIGURED=1) to indicate tiles are configured for use.

FIG. 19(B) illustrates a plurality of registers 1819. As shown each tile has separate registers for its rows and columns. For example, TMM0 rows configuration 1921, TMM0 columns configuration 1923, StartK and StartM are stored in separate registers 1911 and 1913. One or more status registers 1915 are set (e.g., TILES_CONFIGURED=1) to indicate tiles are configured for use.

FIG. 19(C) illustrates a single register 1819. As shown, this register stores tile configurations (rows and columns per tile) 1931, StartK 1933, and StartM 1933 are stored in single register as packed data registers. One or more status registers 1915 are set (e.g., TILES_CONFIGURED=1) to indicate tiles are configured for use.

FIG. 19(D) illustrates a plurality of registers 1819. As shown, a single register stores tile configurations (rows and columns per tile) 1931. StartK and StartM are stored in separate registers 1911 and 1913. One or more status registers 1915 are set (e.g., TILES_CONFIGURED=1) to indicate tiles are configured for use.

Other combinations are contemplated such as combining the start registers into a single register where they are shown separately, etc.

iii. Exemplary Stored Matrix (Tile) Description

FIG. 20 illustrates an embodiment of a description of the matrices (tiles) to be supported. In this example, each field is a byte. In byte[0], a palette ID 2001 is stored. The palette ID is used to index a palette table 1813 which stores, per palette ID, a number of bytes in a tile, and bytes per row of the tiles that are associated with this ID as defined by the configuration. Bytes 1-7 are reserved and are typically zero.

Bytes 8-9 store a value for a “startM” register 2003 and bytes 10-11 store a value for a “startK” register 2005. To support restarting instructions after these events, the instructions store information these registers. The startM indicates a row that should be used for restart. The startK indicates a position in the inner-product for relevant operations. The position in the row (the column) is not needed. Two-dimensional operations like the element-wise addition/subtraction/multiplication only use startM. Three-dimensional operations use values from both startM and startK. Typically, operations that only require startM will zero startK when writing startM.

Any time an interrupted tile instruction is not restarted, in some embodiments, it is the responsibility of software to zero the startM and startK values. For example, unmasked floating point exception handlers might decide to finish the operation in software and change the program counter value to another instruction, usually the next instruction. In this case the software exception handler must zero the startM and startK values in the exception frame presented to it by the operating system before resuming the program. The operating system will subsequently reload those values.

Bytes 16-17 store the number of rows 2013 and columns 2015 for tile 0, bytes 18-19 store the number of rows and columns for tile 1, etc. In other words, each 2 byte group specifies a number of rows and columns for a tile. If a group of 2 bytes is not used to specify tile parameters, they should have the value zero. Specifying tile parameters for more tiles than the implementation limit or the palette limit results in a fault. Unconfigured tiles are set to the INIT state with 0 rows, 0 columns.

Finally, the configuration in memory typically ends with an ending delineation such as all zeros for several consecutive bytes.

iv. Exemplary Format(s)

An embodiment of a format for a TILECONFIG instruction is TILECONFIG Address. In some embodiments, TILECONFIG is the opcode mnemonic of the instruction. Address is a pointer to a matrix (tile) description in memory. In some embodiments, the address field is a R/M value (such as 2446).

In embodiments, encodings of the instruction include a scale-index-base (SIB) type memory addressing operand that indirectly identifies multiple indexed destination locations in memory (e.g., field 2450). In one embodiment, an SIB type memory operand may include an encoding identifying a base address register. The contents of the base address register may represent a base address in memory from which the addresses of the particular destination locations in memory are calculated. For example, the base address may be the address of the first location in a block of potential destination locations for an extended vector instruction. In one embodiment, an SIB type memory operand may include an encoding identifying an index register. Each element of the index register may specify an index or offset value usable to compute, from the base address, an address of a respective destination location within a block of potential destination locations. In one embodiment, an SIB type memory operand may include an encoding specifying a scaling factor to be applied to each index value when computing a respective destination address. For example, if a scaling factor value of four is encoded in the SIB type memory operand, each index value obtained from an element of the index register may be multiplied by four and then added to the base address to compute a destination address.

In one embodiment, an SIB type memory operand of the form vm32{x,y,z} may identify a vector array of memory operands specified using SIB type memory addressing. In this example, the array of memory addresses is specified using a common base register, a constant scaling factor, and a vector index register containing individual elements, each of which is a 32-bit index value. The vector index register may be a 128-bit register (e.g., XMM) register (vm32x), a 256-bit (e.g., YMM) register (vm32y), or a 512-bit (e.g., ZMM) register (vm32z). In another embodiment, an SIB type memory operand of the form vm64{x,y,z} may identify a vector array of memory operands specified using SIB type memory addressing. In this example, the array of memory addresses is specified using a common base register, a constant scaling factor, and a vector index register containing individual elements, each of which is a 64-bit index value. The vector index register may be a 128-bit register (e.g., XMM) register (vm64x), a 256-bit (e.g., YMM) register (vm64y) or a 512-bit (e.g., ZMM) register (vm64z).

v. Exemplary Method(s) of Execution

FIG. 21 illustrates an embodiment of method performed by a processor to process a TILECONFIG instruction.

At 2101, an instruction is fetched. For example, a TILECONFIG instruction is fetched. An embodiment of the TILECONFIG instruction includes fields for an opcode and a memory address operand.

The fetched instruction is decoded at 2103. For example, the fetched TILECONFIG instruction is decoded by decode circuitry such as that detailed herein.

A description found at the memory address of the memory address operand is are retrieved at 2105 and the decoded instruction is scheduled (as needed).

At 2107, the decoded instruction is executed by execution circuitry (hardware) such as that detailed herein. For the TILECONFIG instruction, the execution will cause execution circuitry to configure usage of tiles in a tile configuration (setting the number of rows and columns) and marking that matrix (tile) support is in use (active). For example, configuration one or more registers 1819. Tile support usage (e.g., “TILES_CONFIGURED”) is typically indicated by setting a bit in a status, control, or machine specific register. In particular, instruction execution resources 1811 are configured to use tiles as specified by the retrieved configuration.

In some embodiments, the instruction is committed or retired at 2109.

FIG. 22 illustrates a more detailed description of an execution of a TILECONFIG instruction using memory addressing. Typically, this is performed by execution circuitry such as that detailed above after the description has been retrieved from memory. While not illustrated, in some embodiments, a check is first performed to determine if tiles are supported. Support is usually found by a CPUID check.

At 2201, a determination of if the palette ID is supported is made. For example, does the CPUID state that this ID is supported? If not, then a general protection fault occurs at 2203.

At 2205, a first tile specific grouping is read. For example, the number of rows and columns for tile 0 (T0) is read.

A determination of if the read grouping is valid is made at 2207. For example, if one the number of rows or columns (not both) is set 0, then the grouping is not valid and the configuration halts and tiles are not considered to be in use at 2203. Invalid groups occur, for example, when one of rows or columns (not both) are zero. Additionally, when a value for the number of rows is greater than the maximum of rows supported (this is found by dividing the tile byte size of the palette ID with the number of bytes per row for the palette ID as found in the palette table) as fault occurs. Another potential fault is when there are more names than supported.

If the read grouping is valid, then the tile associated with the read grouping is configured to use the number of rows and columns specified by the grouping in a tile configuration at 2211. The size of the elements in the tile are set by the palette table entry for the palette ID.

A determination of if all tiles of the retrieved configuration have been configured is made at 2213. For example, have all of the possible tile names been processed? In some embodiments, when the rows and columns for a particular tile are both 0, then all tiles have been processed.

When all tiles have not been configured, at 2215, the tile number is incremented such that the next tile in the configuration will be evaluated.

At 2217, the incremented tile's grouping is read. For example, the number of rows and columns for tile 1 (T1) is read. A determination of if the read grouping is valid is made at 2207, etc.

When all tiles have been configured, then the instruction completes at 2209. The tiles will be marked as being in use for matrix operations, for example, by setting an in-use indicator in a register.

vi. Exemplary Pseudocode

FIG. 23 illustrates exemplary pseudocode for an execution of a TILECONFIG instruction.

B. Tile Add

Detailed herein are embodiments of a matrix (tile) addition (“TILEADD”) instruction and its execution. A TILEADD instruction is an improvement to a computer itself as it provides for support to add matrices (tiles) of data values with a single instruction. In particular, the execution of the TILEADD instruction causes elementwise addition of elements of a first source matrix (tile) with corresponding elements of a second source matrix (tile) and storage of the result in corresponding data element positions of a destination matrix (tile). The size of the data values in the source and destination matrices varies depending on the instruction and tile support. Exemplary sizes include, but are not limited to, 16-bit, 32-bit, 64-bit, 128-bit, and 256-bit. In some embodiments, elements of rows of the destination matrix (tile) that do not have corresponding elements in the source matrices (tiles) are zeroed.

i. Exemplary Execution

FIG. 24 illustrates an exemplary execution of a TILEADD instruction. The TILEADD instruction format includes fields for an opcode (e.g., shown as “TADDP{H,S}” in the figure), a destination operand (e.g., shown as “DESTINATION MATRIX (TILE)” in the figure) and two source operands (e.g., shown as “FIRST SOURCE MATRIX (TILE)” and “SECOND SOURCE MATRIX (TILE)” in the figure).

The source and destination matrix (tiles) operand fields represent a first source matrix (tile) 2401, a second source matrix (tile) 2403, and a destination matrix (tile) 2407. As detailed earlier, a matrix (tile) may be stored in a collection of registers, locations in memory (for example, as strided rows), or in other storage accessible to execution circuitry.

As shown, execution circuitry 2405 executes a decoded TADD instruction to perform elementwise addition of the elements of the first source matrix (tile) 2401 and the second source matrix (tile) 2403 and stores the result in corresponding data element positions of the destination matrix (tile) 2407. In some embodiments, a matrix (tile) is configured to use only a subset of the rows and columns possible. For example, a matrix (tile) may have up to 16 rows and columns to use, but only use 4 of each. The configuration of each matrix (tile) is typically done by the execution of a configuration instruction prior to matrix (tile) usage. In this example, there are N columns and M rows possible. Each of the first source matrix (tile) 2401 and the second source matrix (tile) 2403 uses 3 rows and 3 columns. Although the example of FIG. 24 illustrates an example of adding two 3×3 matrices, in general, a TILEADD instruction can operate on any two matrices (tiles) having the same dimensions (that is, source matrices (tiles) having a same number of columns N and a same number of rows M).

In some embodiments, execution circuitry 2405 uses a grid of fused multiply adders (FMAs) to execute a decoded TILEADD instruction by storing the result of adding the two source matrix (tile) operands into corresponding data element positions of the destination matrix (tile) 2407. In particular, the grid of FMAs generates, for each data element position[row, column] of a first source matrix (tile) 2401, a sum of the value at that data element position and the value at a corresponding data element position[row, column] of the second source matrix (tile) 2403. In reference to the example source matrix (tile) operands 2401, 2403, the execution circuitry 2405 generates a sum of the values first source matrix (tile) 2401 position[0,0] and second source matrix (tile) 2403 position[0,0] (A+J) and stores the result in the position[0,0] of the destination matrix (tile) 2407, generates a sum of the values first source matrix (tile) 2401 position[0,1] and second source matrix (tile) 2403 position[0,1] (B+K) and stores the result in the position[0,1] of the destination matrix (tile) 2407, and so forth.

ii. Exemplary Format(s)

An embodiment of a format for a TILEADD instruction is TADDP{H/S} TMM1, TMM2, TMM3. In some embodiments, TADDP{H/S} is the opcode mnemonic of the instruction, where the S or H identifier indicates whether the source matrices (tile) comprise single-precision (PS) or half-precision (PH) floating-point data values. TMM1 is a field for the destination matrix (tile) operand. TMM2 and TMM3 are fields for the matrix (tile) source operands. In some embodiments, the TMM3 field is a R/M value (such as 3646), the TMM1 field is REG 3644, and the data element size is found in 3665.

In embodiments, encodings of the instruction include a scale-index-base (SIB) type memory addressing operand that indirectly identifies multiple indexed destination locations in memory. In one embodiment, an SIB type memory operand may include an encoding identifying a base address register. The contents of the base address register may represent a base address in memory from which the addresses of the particular destination locations in memory are calculated. For example, the base address may be the address of the first location in a block of potential destination locations for an extended vector instruction. In one embodiment, an SIB type memory operand may include an encoding identifying an index register. Each element of the index register may specify an index or offset value usable to compute, from the base address, an address of a respective destination location within a block of potential destination locations. In one embodiment, an SIB type memory operand may include an encoding specifying a scaling factor to be applied to each index value when computing a respective destination address. For example, if a scaling factor value of four is encoded in the SIB type memory operand, each index value obtained from an element of the index register may be multiplied by four and then added to the base address to compute a destination address.

In one embodiment, an SIB type memory operand of the form vm32{x,y,z} may identify a vector array of memory operands specified using SIB type memory addressing. In this example, the array of memory addresses is specified using a common base register, a constant scaling factor, and a vector index register containing individual elements, each of which is a 32-bit index value. The vector index register may be a 128-bit register (e.g., XMM) register (vm32x), a 256-bit (e.g., YMM) register (vm32y), or a 512-bit (e.g., ZMM) register (vm32z). In another embodiment, an SIB type memory operand of the form vm64{x,y,z} may identify a vector array of memory operands specified using SIB type memory addressing. In this example, the array of memory addresses is specified using a common base register, a constant scaling factor, and a vector index register containing individual elements, each of which is a 64-bit index value. The vector index register may be a 128-bit register (e.g., XMM) register (vm64x), a 256-bit (e.g., YMM) register (vm64y) or a 512-bit (e.g., ZMM) register (vm64z).

iii. Exemplary Method(s) of Execution

FIG. 25 illustrates an embodiment of method performed by a processor to process a TILEADD instruction.

At 2501, an instruction is fetched. For example, a TILEADD instruction is fetched. The TILEADD instruction includes fields for an opcode, a first and a second source matrix (tile) operand identifier, and a destination matrix (tile) operand identifier. In some embodiments, the instruction is fetched from an instruction cache. The source operands and destination operand consist of packed data. The opcode of the TILEADD instruction indicates that a sum of the source operands is to be generated. In an embodiment, the opcode further indicates whether the source operands consist of half-precision floating-point values or single-precision floating-point values.

The fetched instruction is decoded at 2503. For example, the fetched TILEADD instruction is decoded by decode circuitry such as that detailed herein.

Data values associated with the source matrix (tile) operands of the decoded instruction are retrieved at 2505 and the decoded instruction is schedule (as needed). For example, when one or more of the source matrix (tile) operands are memory operands, the data from the indicated memory location is retrieved.

At 2507, the decoded instruction is executed by execution circuitry (hardware) such as that detailed herein. For the TILEADD instruction, the execution will cause execution circuitry to perform an elementwise matrix addition operation on the source data. In some embodiments, the execution of a decoded matrix addition operation instruction causes an execution circuit to, for each data element position of the first source matrix operand: add a first data value at that data element position to a second data value at a corresponding data element position of the second source matrix operand, and store a result of the addition into a corresponding data element position of the destination matrix operand.

In some embodiments, a fault is generated when a number of data element rows associated with the first source matrix (tile) operand is different than a number of data element rows associated with the second source matrix (tile) operand or is different than a number of data element rows associated with the destination matrix (tile) operand. Similarly, a fault is generated when a number of data element columns associated with the first source matrix (tile) operand is different than a number of data element columns associated with the second source matrix (tile) operand or is different than a number of data element columns associated with the destination matrix (tile) operand. As described elsewhere herein, the dimensions for each matrix, element size for the data elements each matrix, and other configuration can be set by executing a TILECONFIG instruction.

As described elsewhere herein, successful execution of a TILECONFIG instruction enables subsequent TILE operators and sets a state variable indicating that the corresponding code is in a region with TILES configured. In some embodiments, a fault is generated as part of executing a TILE ADD instruction if the TILES mode is determined to be inactive. For example, the execution circuitry can check whether the state variable set as part of the successful execution of a TILECONFIG instruction indicates that a TILES mode is active.

In some embodiments, the instruction is committed or retired at 2509.

FIG. 26 illustrates an example process describing a method performed by a processor to process a TILEADD instruction. For example, process 2601 illustrates an example method for performing a TILEADD operation when the source matrix (tile) operands contain half-precision elements. Process 2701 illustrates an example method for performing a TILEADD operation when the source matrix (tile) operands contain single-precision elements.

As shown, the process of 2601 determines whether any of the following is true: 1) is a TILES mode not active?; 2) do the destination matrix (tile) operand and the first source matrix (tile) operand have a different number of columns?; 3) do the destination matrix (tile) operand and the second source matrix (tile) operand have a different number of columns?; 4) do the destination matrix (tile) operand and the first source matrix (tile) operand have a different number of row?; 5) do the destination matrix (tile) operand and the second source matrix (tile) operand have a different number of rows?; 6) does the destination matrix (tile) operand have more than a specified maximum number of columns?; 7) does the first source matrix (tile) operand have more than a specified maximum number of columns?; 8) does the second source matrix (tile) operand have more than a specified maximum number of columns? If any of these is true, then a fault is raised.

If none of the conditions above is true, then the execution circuitry writes, for each configured row and column of the identified destination matrix (tile) operand, the sum of corresponding element values from the first source matrix (tile) operand and the second source matrix (tile) operand into a corresponding data element position of the destination matrix (tile) operand. In some embodiments, unconfigured elements of rows of the destination matrix (tile) that do not have corresponding columns in the source matrix (tile) are zeroed.

v. Examples

Example 1 A processor comprising: decode circuitry to decode an instruction having fields for an opcode, a first source matrix operand identifier, a second source matrix operand identifier, and a destination matrix operand identifier; and execution circuitry to execute the decoded instruction to, for each data element position of the identified first source matrix operand: add a first data value at that data element position to a second data value at a corresponding data element position of the identified second source matrix operand, and store a result of the addition into a corresponding data element position of the identified destination matrix operand.

Example 2 The processor of Example 1, wherein the first source matrix operand is a packed data register and the second source matrix operand is a memory location.

Example 3 The processor of Example 1, wherein the first source matrix operand is a packed data register and the second source matrix operand is a packed data register.

Example 4 The processor of any of Examples 1-3, wherein the execution circuitry comprises a plurality of fused-multiply adders.

Example 5 The processor of Example 1, wherein the opcode indicates that each of the first source matrix operand, the second source matrix operand, and the destination matrix operand comprises half-precision floating-point values.

Example 6 The processor of Example 1, wherein the opcode indicates that each of the first source matrix operand, the second source matrix operand, and the destination matrix operand comprises single-precision floating-point values.

Example 7 The processor of any of Examples 1-6, wherein a fault is generated when the first source matrix operand has a different number of data elements than the second source matrix operand.

Example 8 The processor of any of Examples 1-7, wherein a fault is generated when a number of rows associated with the first source matrix operand is different than a number of rows associated with the second source matrix operand.

Example 9 The processor of any of Examples 1-8, wherein a fault is generated when a number of columns associated with the first source matrix operand is different than a number of columns associated with the second source matrix operand.

Example 10 The processor of any of Examples 1-9, wherein the execution circuitry further checks a state variable indicating whether a matrix operations mode is active, and wherein a fault is generated when it is determined that the matrix operations mode is not active.

Example 11 provides a method comprising: decoding an instruction having fields for an opcode, a first source matrix operand identifier, a second source matrix operand identifier, and a destination matrix operand identifier; and executing the decoded instruction to, for each data element position of the identified first source matrix operand: add a first data value at that data element position to a second data value at a corresponding data element position of the identified second source matrix operand, and store a result of the addition into a corresponding data element position of the identified destination matrix operand.

Example 12 The method of Example 11, wherein the first source matrix operand is a packed data register and the second source matrix operand is a memory location.

Example 13 The method of Example 11, wherein the first source matrix operand is a packed data register and the second source matrix operand is a packed data register.

Example 14 The method of any of Examples 11-13, wherein the execution circuitry comprises a plurality of fused-multiply adders.

Example 15 The method of Example 11, wherein the opcode indicates that each of the first source matrix operand, the second source matrix operand, and the destination matrix operand comprises half-precision floating-point values.

Example 16 The method of Example 11, wherein the opcode indicates that each of the first source matrix operand, the second source matrix operand, and the destination matrix operand comprises single-precision floating-point values.

Example 17 The method of any of Examples 11-16, wherein a fault is generated when the first source matrix operand has a different number of data elements than the second source matrix operand.

Example 18 The method of any of Examples 11-17, wherein a fault is generated when a number of rows associated with the first source matrix operand is different than a number of rows associated with the second source matrix operand.

Example 19 The method of Example 11-18, wherein a fault is generated when a number of columns associated with the first source matrix operand is different than a number of columns associated with the second source matrix operand.

Example 20 The method of any of Examples 11-19, wherein executing the decoded instruction includes checking a state variable indicating whether a matrix operations mode is active, and wherein a fault is generated when it is determined that the matrix operations mode is not active.

Example 21 provides a non-transitory machine-readable medium storing an instruction which when executed by a processor causes the processor to perform a method, the method comprising: decoding an instruction having fields for an opcode, a first source matrix operand identifier, a second source matrix operand identifier, and a destination matrix operand identifier, and executing the decoded instruction to, for each data element position of the identified first source matrix operand: add a first data value at that data element position to a second data value at a corresponding data element position of the identified second source matrix operand, and store a result of the addition into a corresponding data element position of the identified destination matrix operand.

Example 22 The non-transitory machine-readable medium of Example 21, wherein the first source matrix operand is a packed data register and the second source matrix operand is a memory location.

Example 23 The non-transitory machine-readable medium of Example 21, wherein the first source matrix operand is a packed data register and the second source matrix operand is a packed data register.

Example 24 The non-transitory machine-readable medium of Example 21, wherein the execution circuitry comprises a plurality of fused-multiply adders.

Example 25 The non-transitory machine-readable medium of Example 21, wherein the opcode indicates that each of the first source matrix operand, the second source matrix operand, and the destination matrix operand comprises half-precision floating-point values.

Example 26 The non-transitory machine-readable medium of Example 21, wherein the opcode indicates that each of the first source matrix operand, the second source matrix operand, and the destination matrix operand contains single-precision floating-point values.

Example 27 The non-transitory machine-readable medium of any of Examples 21-26, wherein a fault is generated when the first source matrix operand has a different number of data elements than the second source matrix operand.

Example 28 The non-transitory machine-readable medium of any of Examples 21-27, wherein a fault is generated when a number of rows associated with the first source matrix operand is different than a number of rows associated with the second source matrix operand.

Example 29 The non-transitory machine-readable medium of any of Examples 21-28, wherein a fault is generated when a number of columns associated with the first source matrix operand is different than a number of columns associated with the second source matrix operand.

Example 30 The non-transitory machine-readable medium of any of Examples 21-29, wherein executing the decoded instruction includes checking a state variable indicating whether a matrix operations mode is active, and wherein a fault is generated when it determines that the matrix operations mode is not active.

Example 31 provides a system comprising: a processor; and an accelerator coupled to the processor, the accelerator including: decode circuitry to decode an instruction having fields for an opcode, a first source matrix operand identifier, a second source matrix operand identifier, and a destination matrix operand identifier; and execution circuitry to execute the decoded instruction to, for each data element position of the identified first source matrix operand: add a first data value at that data element position to a second data value at a corresponding data element position of the identified second source matrix operand, and store a result of the addition into a corresponding data element position of the identified destination matrix operand.

Example 32 The system of Example 31, wherein the first source matrix operand is a packed data register and the second source matrix operand is a memory location.

Example 33 The system of Example 31, wherein the first source matrix operand is a packed data register and the second source matrix operand is a packed data register.

Example 34 The system of any of Examples 31-33, wherein the execution circuitry comprises a plurality of fused-multiply adders.

Example 35 The system of Example 31, wherein the opcode indicates that each of the first source matrix operand, the second source matrix operand, and the destination matrix operand comprises half-precision floating-point values.

Example 36 The system of Example 31, wherein the opcode indicates that each of the first source matrix operand, the second source matrix operand, and the destination matrix operand comprises single-precision floating-point values.

Example 37 The system of any of Examples 31-36, wherein a fault is generated when the first source matrix operand has a different number of data elements than the second source matrix operand.

Example 38 The system of any of Examples 31-37, wherein a fault is generated when a number of rows associated with the first source matrix operand is different than a number of rows associated with the second source matrix operand.

Example 39 The system of any of Examples 31-38, wherein a fault is generated when a number of columns associated with the first source matrix operand is different than a number of columns associated with the second source matrix operand.

Example 40 The system of any of Examples 31-39, wherein the execution circuitry further checks a state variable indicating whether a matrix operations mode is active, and wherein a fault is generated when it is determined that the matrix operations mode is not active.

C. Tile Subtract

Detailed herein are embodiments of a matrix (tile) subtraction (“TILESUB”) instruction and its execution. A TILESUB instruction is an improvement to a computer itself as it provides for support to subtract matrices (tiles) of data values with a single instruction. In particular, the execution of the TILESUB instruction causes elementwise subtraction of elements of a second source matrix (tile) from corresponding elements of a first source matrix (tile) and storage of the result in corresponding data element positions of a destination matrix (tile). The size of the data values in the source and destination matrices varies depending on the instruction and tile support. Exemplary sizes include, but are not limited to, 16-bit, 32-bit, 64-bit, 128-bit, and 256-bit. In some embodiments, elements of rows of the destination matrix (tile) that do not have corresponding elements in the source matrices (tiles) are zeroed.

i. Exemplary Execution

FIG. 28 illustrates an exemplary execution of a TILESUB instruction. The TILESUB instruction format includes fields for an opcode (e.g., shown as “TSUBP(H/S)” in the figure), a destination operand (e.g., shown as “DESTINATION MATRIX (TILE)” in the figure) and two source operands (e.g., shown as “FIRST SOURCE MATRIX (TILE)” and “SECOND SOURCE MATRIX (TILE)” in the figure).

The source and destination matrix (tiles) operand fields represent a first source matrix (tile) 2801, a second source matrix (tile) 2803, and a destination matrix (tile) 2807. As detailed earlier, a matrix (tile) may be stored in a collection of registers, locations in memory (for example, as strided rows), or in other storage accessible to execution circuitry.

As shown, execution circuitry 2805 executes a decoded TSUB instruction to perform elementwise subtraction of the elements of the first source matrix (tile) 2801 and the second source matrix (tile) 2803 and stores the result in corresponding data element positions of the destination matrix (tile) 2807. In some embodiments, a matrix (tile) is configured to use only a subset of the rows and columns possible. For example, a matrix (tile) may have up to 16 rows and columns to use, but only use 4 of each. The configuration of each matrix (tile) is typically done by the execution of a configuration instruction prior to matrix (tile) usage. In this example, there are N columns and M rows possible. Each of the first source matrix (tile) 2801 and the second source matrix (tile) 2803 uses 3 rows and 3 columns. Although the example of FIG. 28 illustrates an example of subtracting two 3×3 matrices, in general, a TILESUB instruction can operate on any two matrices (tiles) having the same dimensions (that is, source matrices (tiles) having a same number of columns N and a same number of rows M).

In some embodiments, execution circuitry 2805 uses a grid of fused multiply adders (FMAs) to execute a decoded TILESUB instruction by storing the result of subtracting the two source matrix (tile) operands into corresponding data element positions of the destination matrix (tile) 2807. In particular, the grid of FMAs generates, for each data element position[row, column] of a first source matrix (tile) 2801, a result value indicating the difference between the value at that data element position and the value at a corresponding data element position[row, column] of the second source matrix (tile) 2803. In reference to the example source matrix (tile) operands 2801 and 2803, the execution circuitry 2805 subtracts the value of the second source matrix (tile) 2803 at position[0,0] from the value of the first source matrix (tile) 2801 at position[0,0] (A-J) and stores the result in the position[0,0] of the destination matrix (tile) 2807, subtracts the value of the second source matrix (tile) 2803 position[0,1] from the value of the first source matrix (tile) 2803 at position[0,1] (B-K) and stores the result in the position[0,1] of the destination matrix (tile) 2807, and so forth.

ii. Exemplary Format(s)

An embodiment of a format for a TILESUB instruction is TSUBP{H/S} TMM1, TMM2, TMM3. In some embodiments, TSUBP{H/S} is the opcode mnemonic of the instruction, where the S or H identifier indicates whether the source matrices (tile) comprise single-precision (PS) or half-precision (PH) floating-point data values. TMM1 is a field for the destination matrix (tile) operand. TMM2 and TMM3 are fields for the matrix (tile) source operands. In some embodiments, the TMM3 field is a R/M value (such as 3646), the TMM1 field is REG 3644, and the data element size is found in 3665.

In embodiments, encodings of the instruction include a scale-index-base (SIB) type memory addressing operand that indirectly identifies multiple indexed destination locations in memory. In one embodiment, an SIB type memory operand may include an encoding identifying a base address register. The contents of the base address register may represent a base address in memory from which the addresses of the particular destination locations in memory are calculated. For example, the base address may be the address of the first location in a block of potential destination locations for an extended vector instruction. In one embodiment, an SIB type memory operand may include an encoding identifying an index register. Each element of the index register may specify an index or offset value usable to compute, from the base address, an address of a respective destination location within a block of potential destination locations. In one embodiment, an SIB type memory operand may include an encoding specifying a scaling factor to be applied to each index value when computing a respective destination address. For example, if a scaling factor value of four is encoded in the SIB type memory operand, each index value obtained from an element of the index register may be multiplied by four and then added to the base address to compute a destination address.

In one embodiment, an SIB type memory operand of the form vm32{x,y,z} may identify a vector array of memory operands specified using SIB type memory addressing. In this example, the array of memory addresses is specified using a common base register, a constant scaling factor, and a vector index register containing individual elements, each of which is a 32-bit index value. The vector index register may be a 128-bit register (e.g., XMM) register (vm32x), a 256-bit (e.g., YMM) register (vm32y), or a 512-bit (e.g., ZMM) register (vm32z). In another embodiment, an SIB type memory operand of the form vm64{x,y,z} may identify a vector array of memory operands specified using SIB type memory addressing. In this example, the array of memory addresses is specified using a common base register, a constant scaling factor, and a vector index register containing individual elements, each of which is a 64-bit index value. The vector index register may be a 128-bit register (e.g., XMM) register (vm64x), a 256-bit (e.g., YMM) register (vm64y) or a 512-bit (e.g., ZMM) register (vm64z).

iii. Exemplary Method(s) of Execution

FIG. 29 illustrates an embodiment of method performed by a processor to process a TILESUB instruction.

At 2901, an instruction is fetched. For example, a TILESUB instruction is fetched. The TILESUB instruction includes fields for an opcode, a first and a second source matrix (tile) operand identifier, and a destination matrix (tile) operand identifier. In some embodiments, the instruction is fetched from an instruction cache. The source operands and destination operand consist of packed data. The opcode of the TILESUB instruction indicates that elementwise subtraction of the source operands is to be generated. In an embodiment, the opcode further indicates whether the source operands consist of half-precision floating-point values or single-precision floating-point values.

The fetched instruction is decoded at 2903. For example, the fetched TILESUB instruction is decoded by decode circuitry such as that detailed herein.

Data values associated with the source matrix (tile) operands of the decoded instruction are retrieved at 2905 and the decoded instruction is schedule (as needed). For example, when one or more of the source matrix (tile) operands are memory operands, the data from the indicated memory location is retrieved.

At 2907, the decoded instruction is executed by execution circuitry (hardware) such as that detailed herein. For the TILESUB instruction, the execution will cause execution circuitry to perform an elementwise matrix subtraction operation on the source data. In some embodiments, the execution of a decoded matrix subtraction operation instruction causes an execution circuit to, for each data element position of the first source matrix operand: subtract from a first data value at that data element position a second data value at a corresponding data element position of the identified second source matrix operand, and store a result of the subtraction into a corresponding data element position of the destination matrix operand.

In some embodiments, a fault is generated when a number of data element rows associated with the first source matrix (tile) operand is different than a number of data element rows associated with the second source matrix (tile) operand or is different than a number of data element rows associated with the destination matrix (tile) operand. Similarly, a fault is generated when a number of data element columns associated with the first source matrix (tile) operand is different than a number of data element columns associated with the second source matrix (tile) operand or is different than a number of data element columns associated with the destination matrix (tile) operand. As described elsewhere herein, the dimensions for each matrix, element size for the data elements each matrix, and other configuration can be set by executing a TILECONFIG instruction.

As described elsewhere herein, successful execution of a TILECONFIG instruction enables subsequent TILE operators and sets a state variable indicating that the corresponding code is in a region with TILES configured. In some embodiments, a fault is generated as part of executing a TILE SUBTRACT instruction if the TILES mode is determined to be inactive. For example, the execution circuitry can check whether the state variable set as part of the successful execution of a TILECONFIG instruction indicates that a TILES mode is active.

In some embodiments, the instruction is committed or retired at 2909.

FIG. 30 illustrates an example process describing a method performed by a processor to process a TILESUB instruction. For example, process 3001 illustrates an example method for performing a TILESUB operation when the source matrix (tile) operands contain half-precision elements. Process 3101 as shown in FIG. 31 illustrates an example method for performing a TILESUB operation when the source matrix (tile) operands contain single-precision elements.

As shown, the process of 3001 determines whether any of the following is true: 1) is a TILES mode not active?; 2) do the destination matrix (tile) operand and the first source matrix (tile) operand have a different number of columns?; 3) do the destination matrix (tile) operand and the second source matrix (tile) operand have a different number of columns?; 4) do the destination matrix (tile) operand and the first source matrix (tile) operand have a different number of row?; 5) do the destination matrix (tile) operand and the second source matrix (tile) operand have a different number of rows?; 6) does the destination matrix (tile) operand have more than a specified maximum number of columns?; 7) does the first source matrix (tile) operand have more than a specified maximum number of columns?; 8) does the second source matrix (tile) operand have more than a specified maximum number of columns? If any of these is true, then a fault is raised.

If none of the conditions above is true, then the execution circuitry writes, for each configured row and column of the identified destination matrix (tile) operand, the sum of corresponding element values from the first source matrix (tile) operand and the second source matrix (tile) operand into a corresponding data element position of the destination matrix (tile) operand. In some embodiments, unconfigured elements of rows of the destination matrix (tile) that do not have corresponding columns in the source matrix (tile) are zeroed.

iv. Examples

Example 1 A processor comprising: decode circuitry to decode an instruction having fields for an opcode, a first source matrix operand identifier, a second source matrix operand identifier, and a destination matrix operand identifier; and execution circuitry to execute the decoded instruction to, for each data element position of the identified first source matrix operand: subtract from a first data value at that data element position a second data value at a corresponding data element position of the identified second source matrix operand, and store a result of the subtraction into a corresponding data element position of the identified destination matrix operand.