Patent Grant
9575765
The field of invention relates generally to processor architecture, and, more specifically, to techniques for coalescing gather scatter operations.
In order to fully utilize the single instruction, multiple data (SIMD) processor, gather instructions are used to read a set of (possibly) non-contiguous source data elements from memory and pack them together, typically into a single register. Scatter instructions do the reverse. In some instances, the data elements in memory are known to be contiguous to each other. Unfortunately, conventional gather and scatter instructions do not leverage off this known information, thus reducing the efficiency of the SIMD processor.
Embodiments of the invention are illustrated by way of example and not limitation in the figures of the accompanying drawings in which like references indicate similar elements.
Various embodiments and aspects of the inventions will be described with reference to details discussed below, and the accompanying drawings will illustrate the various embodiments. The following description and drawings are illustrative of the invention and are not to be construed as limiting the invention. Numerous specific details are described to provide a thorough understanding of various embodiments of the present invention. However, in certain instances, well-known or conventional details are not described in order to provide a concise discussion of embodiments of the present inventions.
Reference in the specification to “one embodiment” or “an embodiment” means that a particular feature, structure, or characteristic described in conjunction with the embodiment can be included in at least one embodiment of the invention. The appearances of the phrase “in one embodiment” in various places in the specification do not necessarily all refer to the same embodiment.
Referring now to
Referring still to
According to some embodiments, a new instruction set architecture (ISA) is utilized to perform coalesced gathering of contiguous data elements from memory and storing them into a set of multiple destination registers. The new ISA is also utilized to perform coalesced scattering of data elements from multiple source operands (e.g., registers, memory locations) by storing them in contiguous data elements in memory. A new set of processor instructions is used to implement the coalesced gathering/scattering with significant performance improvement with respect to existing processor instructions. The ISA is defined to operate on 128-bit SIMD registers (e.g., XMM registers), 256-bit SIMD registers (e.g., YMM registers), or 512-bit SIMD registers (e.g., ZMM registers). The SIMD register width discussed above are only for illustrative purposes. It will be appreciated that other the ISA may be defined to operate with other SIMD register widths. Embodiments of coalesced gathering techniques include reading from memory, using a single memory access, contiguous data elements and storing them in multiple destination storage locations. Embodiments of coalesced scattering techniques include reading data elements from multiple source registers, and storing them in memory contiguously in a single memory access. In the description provided herein, contiguous data elements shall mean that the data elements are located adjacent to each other in memory. Thus, the location in memory corresponding to the end of one data element is adjacent to the location in memory corresponding to the start of another data element.
In one embodiment, processor 200 includes, but is not limited to, instruction decoder 202 to receive and decode instruction 210. Instruction decoder 202 may generate and output one or more micro-operations, micro-code, entry points, microinstructions, other instructions, or other control signals, which reflect, or are derived from, instruction 210. Instruction decoder 202 may be implemented using various different mechanisms. Examples of suitable mechanisms include, but are not limited to, microcode read only memories (ROMs), look-up tables, hardware implementations, programmable logic arrays (PLAs), and the like.
Execution units 204, which may include an arithmetic logic unit (ALU), or another type of logic unit capable of performing operations based on instructions. As a result of instruction decoder 202 decoding the instruction 210, execution unit 204 may receive one or more micro-operations, micro-code entry points, microinstructions, other instructions, or other control signals, which reflect, or are derived from, the instructions. Execution unit 204 may be operable as a result of instructions indicating one or more source operands (SRC) and to store a result in one or more destination operands (DEST) of a register set indicated by the instructions. Execution unit 204 may include circuitry or other execution logic (e.g., software combined with hardware and/or firmware) operable to execute instructions or other control signals derived from the instructions and perform an operation accordingly.
In one embodiment, instruction 210 may implicitly indicate and/or explicitly specify (e.g., through one or more dedicated fields or sets of bits) the source and destination operands. Examples of suitable sources and/or destination of the operands include registers, memory, immediate of the instruction, and combinations thereof. In various embodiments, the source and destination operands may be 8-bit, 16-bit, 32-bit, 64-bit, 128-bit, 256-bit, or 512-bit operands, although this is not required.
Some or all of the source and destination operands may be stored in storage resources 206 such as registers of a register set or memory. A register set may be part of a register file, along with potentially other registers, such as status registers, flag registers, etc. A register may be a storage location or device that may be used to store data. The register set may often be physically located on die with the execution unit(s). The registers may be visible from the outside of the processor or from a programmer's perspective. For example, instructions may specify operands stored in the registers. Various different types of registers are suitable, as long as they are capable of storing and providing data as described herein. The registers may or may not be renamed Examples of suitable registers include, but are not limited to, dedicated physical registers, dynamically allocated physical registers using register renaming, combinations of dedicated and dynamically allocated physical registers, etc. Alternatively, one or more of the source and destination operands may be stored in a storage location other than a register, such as, for example, a location in system memory.
According to one embodiment, execution unit 204 includes gather/scatter unit 208 that executes gather/scatter instructions that have been decoded by instruction decoder 202. Below are embodiments of gathering and scattering instructions, which when executed by gather/scatter unit 208, improve efficiency of a SIMD system by taking advantage of the fact that data elements are contiguously located in memory.
In one embodiment, the gather instruction is a coalesced gather instruction. The execution of this instruction by gather/scatter unit 208 stores contiguous data elements from memory into multiple destination operands. For example, in some embodiments, up to sixteen 32-bit or eight 64-bit floating-point data elements are conditionally packed into destination operands such as XMM, YMM, or ZMM registers.
The contiguous memory data elements to be loaded into destination operands are specified via a type of SIB (scale, index, and base) addressing. The coalesced gather instruction also includes a writemask. In some embodiments that use a dedicated mask register such as a “k” writemask (detailed later), the memory data elements will be loaded when their corresponding writemask bit indicates that they should be (for example, in some embodiments if the bit is a “1”). If a memory data element's corresponding writemask bit is not set, the corresponding data element of the destination operand (e.g., an XMM, YMM, or ZMM register) is left unchanged.
In one embodiment, the execution of a coalesced gather instruction will result in the entire writemask register being set to zero unless there is an exception. However, in some embodiments the instruction is suspended by an exception if at least one element has already been gathered (i.e., if the exception is triggered by an element other than the least significant one with its writemask bit set). When this happens the destination register and the writemask register are partially updated (those elements that have been gathered are placed into the destination register and have their mask bits set to zero). If any traps or interrupts are pending from already gathered elements, they may be delivered in lieu of the exception and the EFLAGS resume flag or equivalent is set to one so an instruction breakpoint is not re-triggered when the instruction is continued.
In some embodiments with 128-bit size destination registers, the instruction will gather up to four single-precision floating point values or two double-precision floating point values per destination register. In some embodiments with 256-bit size destination registers, the instruction will gather up to eight single-precision floating point values or four double-precision floating point values per destination register. In some embodiments with 512-bit size destination registers, the instruction will gather up to sixteen single-precision floating point values or eight double-precision floating point values per destination register.
In some embodiments, if the mask and destination registers are the same, this instruction delivers a general protection (GP) fault. Typically, the data element values may be read from memory in any order. However, faults are delivered in a right-to-left manner. That is, if a fault is triggered by an element and delivered, all elements closer to the least significant bit (LSB) of the destination XMM, YMM, or ZMM will be completed (and non-faulting). Individual elements closer to the MSB may or may not be completed. If a given element triggers multiple faults, they are delivered in the conventional order. A given implementation of this instruction is repeatable—given the same input values and architectural state, the same set of elements to the left of the faulting one will be gathered.
The coalesced gather instruction may be implemented in several formats. In one embodiment, the coalesced gather instruction is defined as follows:
VGATHERQ4PD zmm3:zmm5:zmm6:zmm0{k1}, [rax+zmm9]//Format 0
where zmm3, zmm5, zmm6, and zmm0 are destination vector register operands (such as a 128-, 256-, 512-bit register, etc.), k1 is a writemask operand (such as a 16-bit register examples of which are detailed later), rax is the base address, and zmm9 is a vector/array of indices. Note that the above format is described for illustration purposes only; other formats or orders of the operands may be implemented. In one embodiment, the base and a value stored in an index of the vector of indices are used to generate a memory address corresponding to the starting location of a block of contiguous data elements which are read and stored into the corresponding data elements (i.e., entries) of the destination operands. In some embodiments, the writemask is also of a different size (8 bits, 32 bits, etc.). Additionally, in some embodiments, not all bits of the writemask are utilized by the instruction.
In one embodiment, in the above format 0 of the instruction, the first destination operand is zmm3, the second destination operand is zmm5, the third destination operand is zmm6, and the fourth destination operand is zmm0 In another embodiment, the order of the operands is the reverse. In one embodiment, the order of these operands explicitly indicates the order in which contiguous data elements in memory will be loaded into the destination operands. Thus, in the above format 0 example, assuming the writemask indicates all data elements are to be updated (discussed in further details below), and assuming further that zmm3 is the first operand, data element at memory location “rax+zmm9[0]” is stored into zmm3[0]. The next three contiguous data elements, i.e., data elements at memory location “rax+zmm9[0]+size of (data element)”, at “rax+zmm9[0]+2*size of(data element))”, and at “rax+zmm9[0]+(3*size of(data element))” are stored into the first data element of each of the subsequent destination operands, i.e., zmm5[0], zmm6[0], and zmm0[0], respectively. The second data element of each destination operand will be updated with contiguous data elements in memory using the same addressing scheme, e.g., zmm3[1] will be updated with the data element at memory location “rax+zmm9[1]”, and zmm5[1], zmm6[1], zmm0[1] will be loaded with the next three contiguous data elements in memory.
VGATHERQ4PD is the instruction's opcode. Typically, each operand is explicitly defined in the instruction. The size of the data elements may be defined in the “suffix” of the instruction. For instance, the suffix “PD” in the instruction VGATHERQ4PD may indicate that the data element is a double precision (i.e., 64 bits). In most embodiments, data elements are either 32 or 64 bits. If the data elements are 32 bits in size, and the operands are 512 bits in size, then there are sixteen (16) data elements per operand. In some embodiments, the number of data elements per operand implicitly indicates the number of indices that are present in the vector of indices (e.g., zmm9 in the above example.) In some embodiments, the number of operands is also explicitly defined in the instruction. For instance, in the above example, the “4” preceding the “PD” suffix may indicate that the instruction is coalescing four adjacent gather operations, i.e., the execution of the instruction results in writing four contiguous data elements from memory into corresponding data elements of four destination operands (e.g., zmm3, zmm5, zmm6, and zmm0 in the above example). In one embodiment, the block of contiguous data elements are read from memory in a single memory access. In one embodiment, the block of data elements are stored into all the destination operands in a single cycle.
In another embodiment, the coalesced gather instruction is defined as follows:
VGATHERQ4PD zmm3-zmm0 {k1}, [rax+zmm9].//Format 1
The execution of the coalesced gather instruction having format 1 causes operations similar to those discussed above in the text relating to format 0 to be executed. The difference between format 0 and format 1 is that with format 1, the destination registers are specified as a range of registers. In the above example of format 1, the range of destination registers are bound by zmm3 and zmm0 Thus, implicitly, the destination registers are zmm3, zmm2, zmm1, and zmm0, where zmm2 and zmm1 are implied by the fact that the instruction explicitly indicates four destination registers are to be packed with data elements from memory. Note that in this embodiment, although the choice of the first destination register may be freely specified, it is a syntax error to specify a range of destination registers that is inconsistent with the number of destination registers explicitly indicated by the instruction.
In another embodiment, the coalesced gather instruction is defined as follows:
VGATHERQ4PD zmm3-zmm0{k1}, [rax+zmm9]//Format 2
The execution of the coalesced gather instruction having format 2 causes operations similar to those discussed above in the text relating to format 0 to be executed. The difference between format 0 and format 2 is that with format 2, the destination registers are fixed. Thus, for instance, in the above example, the destination registers are fixed to zmm3, zmm2, zmm1, and zmm0 because the instruction explicitly indicates four destination registers are to be packed with data elements from memory. Note that in this embodiment, it is a syntax error to specify any destination registers other than “zmm3-zmm0” and this is specified only as an aid to readability. Although in the above example, the registers are fixed to “zmm3-zmm0”, it will be appreciated that the registers may be fixed to a range of other registers, e.g., “zmm4-zmm1”, or “zmm5-zmm2”, etc., or fixed to a set of non-contiguous registers.
In one embodiment, the data elements are fetched from memory and stored in temporary vector registers in a manner similar to that discussed above, prior to being stored in the destination registers.
Referring now to
The execution of the coalesced gather instruction causes gather/scatter unit 208 to generate a first memory address and determine if read 0 should be carried out. In one embodiment, the first address is the base address plus the value stored in the first index of the array of indices (i.e., “rax+zmm8[0]”), which points to a memory location of the first data element in memory to be stored in the first data element of the first destination operand (zmm2[0]).
In one embodiment, gather/scatter unit 208 determines if data elements should be read from memory and stored in the corresponding data elements of the destination operands according to the writemask bit values. In this illustration, the first (LSB) bit of the writemask, i.e., k1[0] is “1”, which in one embodiment, indicates that the first (LSB) data element of each destination operands should be updated. As a result, the data element at memory location “rax+zmm8[0]” having the value “2” and the next two contiguous data elements having the values “1” and “0”, are read from memory in a single memory access. In one embodiment, the data elements {2, 1, 0} are stored, in a single cycle, into the zmm2[0], zmm1[0], and zmm0[0], respectively.
Similarly, gather/scatter unit 208 generates a second address pointing to the memory location “rax+zmm8[1]” and determines if read 1 should be carried out. Like k1[0], the writemask bit k1[1] is also set to “1”, thus contiguous data elements {12, 11, 10} are fetched from memory starting at location “rax+zmm8[1]” and stored into zmm2[1], zmm1[1], and zmm0[1], respectively.
Gather/scatter unit 208 skips read 2 in this example because unlike k1[0] and k1[1], writemask bit k1[2] is set to “0”, which in one embodiment, indicates that the third data element of the destination operands should not be updated. As a result, zmm2[2], zmm1[2], and zmm0[2] remain unchanged, as indicated by the label “x” in
Gather/scatter unit 208 performs the same logic as discussed above and determines that read 3, read 4, and read 6 should be skipped because k1[3], k1[4], and k1[6] are all set to “0”. Moreover, gather/scatter unit 208 determines that read 5 should be performed because k1[5] is set to “1”, and fetches the contiguous data elements {52, 51, 50} from memory starting at the address “rax+zmm8[5]” and stores them in zmm2[5], zmm1[5], and zmm0[5], respectively. Likewise, read 7 is performed because k1[7] is set to “1”, and contiguous data elements {72, 71, 70} are fetched from memory starting at location “rax+zmm8[7]” and stored into zmm2[7], zmm1[7], and zmm0[7], respectively.
At block 510, in response to the first instruction, contiguous first and second data elements are read from memory based on the memory address indicated by the third operand. In one embodiment, the contiguous data elements are read from memory using a single memory access.
At block 515, the first instruction is executed to store contiguous first and second data elements from memory into a first entry of the first and second storage location, respectively. In one embodiment, the contiguous data elements are stored into the first and second storage locations in a single cycle. In one embodiment, the data elements are stored into temporary vector registers prior to being stored into destination operands.
At block 610, it is determined, according to a value of the first writemask bit, if the data elements are to be stored into the first entry of the first and second storage location. In one embodiment, the first writemask bit is the LSB bit of the writemask. When the writemask bit does not indicate that the data elements should be stored in the first and second storage location, then the first entry of the first and second storage location are left unchanged at block 630, and the processing is completed. In one embodiment, a writemask bit having a value of “0” indicates that the data elements should not be stored in the first and second storage location. In another embodiment, the opposite convention is be used.
At block 615, when the value stored in the first mask bit of the writemask indicates that the data elements should be stored in the first entry of the first and second storage location, then data elements are read from memory. In one embodiment, the first data element is located at the first address, and the second data element is contiguously located next to the first data element. In one embodiment, the data elements are read from memory in a single memory access.
At block 620, the first and second data elements are stored into the first and second storage location. In one embodiment, the storage location is an array of entries (e.g., an array of data elements), and the first entry of the storage location is the LSB entry. In one embodiment, the data elements are stored into the first entry of the first and second storage location in a single cycle. In one embodiment, the first and second data elements are stored in vector registers prior to being stored into the first and second storage location.
At block 625, the first writemask bit is cleared to indicate that the corresponding block of data elements have been fetched and stored in the first and second storage location, and the process is completed.
Referring back to
The source operand data elements to be loaded into memory are specified via a type of SIB (scale, index, and base) addressing. The coalesced scatter instruction also includes a writemask. In some embodiments that use a dedicated mask register such as a “k” writemask, the source operand data elements will be loaded into memory when their corresponding writemask bit indicates that they should be (for example, in some embodiments if the bit is a “1”). If a data element's corresponding writemask bit is not set, the corresponding data element in memory is left unchanged.
In some embodiments with 128-bit size source registers, the instruction will scatter up to four single-precision floating point values or two double-precision floating point values per source register. In some embodiments with 256-bit size source registers, the instruction will scatter up to eight single-precision floating point values or four double-precision floating point values per source register. In some embodiments with 512-bit size source registers, the instruction will scatter up to sixteen single-precision floating point values or eight double-precision floating point values per source register.
The coalesced scatter instruction may be implemented in several formats. In one embodiment, the coalesced scatter instruction is defined as follows:
VSCATTERQ4PD[rax+zmm9] {k1}, zmm3:zmm5:zmm6:zmm0//Format 3
where zmm3, zmm5, zmm6, and zmm0 are source vector register operands (such as a 128-, 256-, 512-bit register, etc.), k1 is a writemask operand (such as a 16-bit register examples of which are detailed later), rax is the base address, and zmm9 is a vector/array of indices. In one embodiment, the base and a value stored in an index of the vector of indices are used to generate a memory destination address where the first data element of the first source operand will be stored. In some embodiments, the writemask is also of a different size (8 bits, 32 bits, etc.). Additionally, in some embodiments, not all bits of the writemask are utilized by the instruction.
In one embodiment, in the above format 3 of the coalesced scatter instruction, the first source operand is zmm3, the second source operand is zmm5, the third source operand is zmm6, and the fourth source operand is zmm0. In another embodiment, the order of the source operands is the reverse, e.g., zmm3 is the fourth operand and zmm0 is the first operand. In one embodiment, the order of these operands explicitly indicates the order in which data elements from each source operand will be stored into contiguous memory. Thus, in the above format 3 example, assuming the writemask indicates all data elements are to be updated (discussed in further details below) and assuming further that the first source operand is zmm3, source operand data element zmm3[0] is stored as the first data element in the contiguous block of memory starting at memory location “rax+zmm9[0]”. The data elements of the next three source operands, i.e., zmm5[0], zmm6[0], and zmm0[0] are stored in contiguous memory location, i.e., zmm5[0] is stored at “rax+zmm9[0]+size of(data element)”, zmm6[0] is stored at “rax+zmm9[0]+(2*size of(data element))”, and zmm0[0] is stored at “rax+zmm9[0]+(3*size of(data element))”. The second data element of each source operand will be stored in contiguous memory location starting at location “rax+zmm9[1]”, similar to how the first data elements of the source operands are stored. Thus, zmm3[1] is stored at “rax+zmm9[1]”; zmm5[1] is stored at “rax+zmm9[1]+size of(data element)”; zmm6[1] is stored at “rax+zmm9[1]+(2*size of(data element))”; and zmm0[1] is stored at “rax+zmm9[1]+(3*size of(data element))”. The remaining data elements of the source operands are stored into contiguous blocks of memory using the same logic.
VSCATTERQ4PD is the instruction's opcode. Typically, each operand is explicitly defined in the instruction. The size of the data elements may be defined in the “suffix” of the instruction. For instance, the suffix “PD” in the instruction VSCATTERQ4PD may indicate that the data element is a double precision (i.e., 64 bits). In most embodiments, data elements are either 32 or 64 bits. If the data elements are 32 bits in size, and the operands are 512 bits in size, then there are sixteen (16) data elements per operand. In some embodiments, the number of data elements per operand implicitly indicates the number of indices that are present in the vector of indices (e.g., zmm9 in the above example.) In one embodiment, the number of operands is also explicitly defined in the instruction. For instance, in the above example, the “4” preceding the “PD” suffix may indicate that the instruction is coalescing four adjacent scatter operations, i.e., the execution of the instruction results in writing data elements from four source operands into contiguous memory blocks of four data elements. In one embodiment, the source operand data elements are written to each contiguous memory block in a single memory access.
In another embodiment, the coalesced scatter instruction is defined as follows:
VSCATTERQ4PD [rax+zmm9] {k1}, zmm3-zmm0//Format 4
The execution of the coalesced scatter instruction having format 4 causes operations similar to those discussed above in the text relating to format 3 to be executed. The difference between format 3 and format 4 is that with format 4, the source operand registers are specified as a range of registers. In the above example of format 4, the range of source registers are bound by zmm3 and zmm0. Thus, implicitly, the source registers are zmm3, zmm2, zmm1, and zmm0, where zmm2 and zmm1 are implied by the fact that the instruction explicitly indicates data elements from four source registers are to be packed into memory. Note that in this embodiment, although the choice of the first source register may be freely specified, it is a syntax error to specify a range of source registers that is inconsistent with the number of source registers explicitly indicated by the instruction.
In another embodiment, the coalesced scatter instruction is defined as follows:
VSCATTERQ4PD[rax+zmm9] {k1}, zmm3-zmm0.//Format 5
The execution of the coalesced scatter instruction having format 5 causes operations similar to those discussed above in the text relating to format 3 to be executed. The difference between format 3 and format 5 is that with format 5, the source registers are fixed. Thus, for instance, in the above example, the source registers are fixed to zmm3, zmm2, zmm1, and zmm0 because the instruction explicitly indicates data elements from four source registers are to be packed into memory. Note that in this embodiment, it is a syntax error to specify any source registers other than “zmm3-zmm0” and this is specified only as an aid to readability. Although in the above example, the registers are fixed to “zmm3-zmm0”, it will be appreciated that the registers may be fixed to a range of other registers, e.g., “zmm4-zmm1”, or “zmm5-zmm2”, etc.
The execution of the coalesced scatter instruction causes gather/scatter unit 208 to generate a first address and determine if write 0 should be carried out. In one embodiment, the first address is the base address plus the value stored in the first index of the array of indices (i.e., “rax+zmm8[0]”), which points to a memory location of the start of the block of contiguous memory, where the first data element of the first source operand register (zmm2[0]) is to be stored.
In one embodiment, gather/scatter unit 208 determines if source operand data elements should be stored into memory according to the writemask bit values. In this illustration, the first (LSB) bit of the writemask, i.e., k1[0], is “1”, which in one embodiment, indicates that the first (LSB) data element of each source operands should be packed and stored into contiguous memory. As a result, source operand data elements {2, 1, 0} of zmm2[0], zmm1[0], and zmm0[0], respectively, are packed and stored as a block of contiguous memory, starting at the memory location “rax+zmm8[0]”.
Similarly, gather/scatter unit 208 generates a second address pointing to the memory location “rax+zmm8[1]” and determines if write 1 should be carried out. Like k1[0], the writemask bit k1[1] is also set to “1”, thus data elements {12, 11, 10} of zmm2[1], zmm1[1], and zmm0[0], respectively, are packed and stored into contiguous block of memory starting at the memory location “rax+zmm8[1]”.
Gather/scatter unit 208 skips write 2 in this example because unlike k1[0] and k1[1], writemask bit k1[2] is set to “0”, which in one embodiment, indicates that the third data element of the source operands should not be stored into memory. As result, zmm2[2], zmm1[2], and zmm0[2] are not written to memory.
Gather/scatter unit 208 performs the same logic as discussed above and determines that write 3, write 4, and write 6 should be skipped because k1[3], k1[4], and k1[6] are all set to “0”. Moreover, gather/scatter unit 208 determines that write 5 should be performed because k1[5] is set to “1”, and stores data elements {52, 51, 50} of source operands zmm2[5], zmm1[5], and zmm0[5] into contiguous memory starting at the address “rax+zmm8[5]”. Likewise, write 7 is performed because k1[7] is set to “1”, and data elements {72, 71, 70} of source operands zmm2[7], zmm1[7], and zmm0[7] are stored into contiguous memory starting at location “rax+zmm8[7]”.
In the above discussion, adjacent gather/scatter instructions are coalesced by a new ISA. It will be appreciated, however, that adjacent gather/scatter operations may also be coalesced using the current ISA by combining them “behind the scene”. For example, the three gather instructions:
vgatherqpd zmm0{k1}, [rax+zmm8+0]
vgatherqpd zmm1{k1}, [rax+zmm8+8]
vgatherqpd zmm2{k1}, [rax+zmm8+16]
may be performed by the current SIMD hardware as a single coalesced gather instruction. In one embodiment, in order to be combined as a coalesced gather instruction, the three gather instructions above must have the same operands, i.e., base, scale, index and writemask. Moreover, the instructions must have the right offsets/displacements. For example, the offset of each gather instruction must be a multiple of the size of the data element, so that each data element of an instruction is contiguously located in memory next to the data element of the preceding instruction. In one embodiment, when the above gather instructions are received, gather/scatter unit 208 assumes that they can be coalesced, and issues the combined reads, and then before retiring the instructions, checks that the instructions can be coalesced based on their operands. If not, the results are discarded and the instructions are re-executed as separate gathers.
In one embodiment, the following scatter operations may be similarly coalesced “behind the scene”:
vscatterqpd [rax+zmm8+0] {k1}, zmm0
vscatterqpd [rax+zmm8+8] {k1}, zmm1
vscatterqpd [rax+zmm8+16] {k1}, zmm2.
In another embodiment, adjacent gather/scatter operations are coalesced using the current ISA by adding a prefix to the instructions to strongly hint that they will be fused/coalesced. For example, the gather instructions:
repvgatherqpd zmm0{k1}, [rax+zmm8+0]
repvgatherqpd zmm1 {k1}, [rax+zmm8+8]
vgatherqpd zmm2 {k1}, [rax+zmm8+16]
may be performed by the SIMD hardware as a single coalesced gather instruction. In this embodiment, the prefix “rep” assures the hardware that there are further gathers that can be coalesced coming along, and that it should buffer the first few until the last gather (which has no prefix) arrives. Similarly, the following strongly hinted scatter instructions may be coalesced:
repvscatterqpd [rax+zmm8+0] {k1}, zmm0
repvscatterqpd [rax+zmm8+8] {k1}, zmm1
vscatterqpd [rax+zmm8+16] {k1}, zmm2.
Referring now to
In another embodiment, adjacent gather and scatter operations are coalesced by transposing partial gather/scatter instructions. As discussed above, to gather a 2-element structure from eight indices currently requires two gather instructions:
vgatherqpd zmm0 {k1}, [rax+zmm8+0]
vgatherqpd zmm1 {k2}, [rax+zmm8+8].
Assuming k1=k2=all “1s”, each of these instructions looks at all eight indices in zmm8 and performs a single 8-byte load. This results in sixteen cache line accesses, which is twice as many as required. In the following discussion, the x,y,z,w naming convention for data elements will be used, and the shorthand “x0” means “the double-precision value at address “rax+zmm8[0]+0”. Similarly, “y3” means “the double-precision value at address rax+zmm8[3]+8”. Given this naming convention, the execution of the above gather instructions produces the following results:
In one embodiment of the invention, a “partial” gather is performed that only uses some of the indices, but in return can load more data per index. This can be illustrated as a pair of partial gather instructions:
vgatherp0123qpd zmm0{k1}, [rax+zmm8+0]
vgatherp4567qpd zmm1{k2}, [rax+zmm8+0].
The “0123” part of the first instruction indicates to gather/scatter unit 208 that the instruction only uses the first four indices of zmm8 and the first four bits of the writemask k1. Similarly, the “4567” of the second instruction indicates that it only uses the second four indices and writemask bits. Thus, the results are:
The reason for the odd ordering of the results will be explained in further details below.
The above is a description of a 2-element structure. The equivalent for a 4-element structure is similar:
vgatherp01qpd zmm0 {k1}, [rax+zmm8+0]
vgatherp23qpd zmm1{k2}, [rax+zmm8+0]
vgatherp34qpd zmm2{k1}, [rax+zmm8+0]
vgatherp67qpd zmm3{k2}, [rax+zmm8+0].
Each of the above partial gather instructions performs only two reads, but each read is 256 bits in size. The results look like this:
One of the advantages of the partial gather operations above is that they reduce the number of memory accesses. However, the disadvantage is that the data elements written to the vector registers are not in the format that is useful to standard vector arithmetic. In one embodiment, this formatting irregularity can be resolved by the existing shuffle/permute operations in the ISA. It is clearly possible to perform the 2-element transpose with 4 VPERMD instructions with the appropriate writemasks and the 4-element transpose in 16 VPERMD instructions with writemasks. Alternatively the newer VPERMI2W instruction can permute data from two register sources, which can be used to halve the number of instructions required.
Even when using these existing permutation instructions, the new sequence of gathers can outperform the existing sequences because of the significant reduction in memory accesses.
In one embodiment, new special-purpose instructions are used to perform the transpose in only two or four instructions, by taking advantage of the fact that the VPUs are constructed as four “banks” of ALU and register file blocks, each bank handling 128 bits of the result. This means that each bank can read a different source register to the adjacent bank, allowing the gather/scatter unit 208 to read parts of up to four registers while only using a single read port per bank. This allows a single transpose instruction to read data from all four registers, and then send the combined 512-bit temporary result to the shuffle unit (called “GENMUX”) to reorder the data into the correct order.
In one embodiment, these operations are specified as multiple instructions with hard-coded selection and permutation controls. In another embodiment, they are specified more flexibly with the controls coming from either immediate values or from registers. Here, for simplicity we show them as hardcoded discrete instructions.
Using these transpose instructions allows the full gather+transpose operation to be performed very quickly. The 2-component version requires two gather and two transpose instructions:
vgatherp0123qpd zmm0{k1}, [rax+zmm8+0]
vgatherp4567qpd zmm1{k1}, [rax+zmm8+0]
vtranspose0123pd zmm10, zmm0, zmm1
vtranspose4567pd zmm11, zmm0, zmm1.
The 4-component version requires four gather and four transpose instructions:
vgatherp01qpd zmm0{k1}, [rax+zmm8+0]
vgatherp23qpd zmm1{k1}, [rax+zmm8+0]
vgatherp45qpd zmm2{k1}, [rax+zmm8+0]
vgatherp67qpd zmm3{k1}, [rax+zmm8+0]
vtranspose01pd zmm10, zmm0, zmm1, zmm2, zmm3
vtranspose23pd zmm11, zmm0, zmm1, zmm2, zmm3
vtranspose45pd zmm12, zmm0, zmm1, zmm2, zmm3
vtranspose67pd zmm13, zmm0, zmm1, zmm2, zmm3.
An instruction set, or instruction set architecture (ISA), is the part of the computer architecture related to programming, and may include the native data types, instructions, register architecture, addressing modes, memory architecture, interrupt and exception handling, and external input and output (I/O). The term instruction generally refers herein to macro-instructions—that is instructions that are provided to the processor (or instruction converter that translates (e.g., using static binary translation, dynamic binary translation including dynamic compilation), morphs, emulates, or otherwise converts an instruction to one or more other instructions to be processed by the processor) for execution—as opposed to micro-instructions or micro-operations (micro-ops)—that is the result of a processor's decoder decoding macro-instructions.
The ISA is distinguished from the microarchitecture, which is the internal design of the processor implementing the instruction set. Processors with different microarchitectures can share a common instruction set. For example, Intel® Pentium 4 processors, Intel® Core™ processors, and processors from Advanced Micro Devices, Inc. of Sunnyvale Calif. implement nearly identical versions of the x86 instruction set (with some extensions that have been added with newer versions), but have different internal designs. For example, the same register architecture of the ISA may be implemented in different ways in different microarchitectures using well-known techniques, including dedicated physical registers, one or more dynamically allocated physical registers using a register renaming mechanism (e.g., the use of a Register Alias Table (RAT), a Reorder Buffer (ROB), and a retirement register file; the use of multiple maps and a pool of registers), etc. Unless otherwise specified, the phrases register architecture, register file, and register are used herein to refer to that which is visible to the software/programmer and the manner in which instructions specify registers. Where a specificity is desired, the adjective logical, architectural, or software visible will be used to indicate registers/files in the register architecture, while different adjectives will be used to designation registers in a given microarchitecture (e.g., physical register, reorder buffer, retirement register, register pool).
An instruction set includes one or more instruction formats. A given instruction format defines various fields (number of bits, location of bits) to specify, among other things, the operation to be performed (opcode) and the operand(s) on which that operation is to be performed. Some instruction formats are further broken down though the definition of instruction templates (or subformats). For example, the instruction templates of a given instruction format may be defined to have different subsets of the instruction format's fields (the included fields are typically in the same order, but at least some have different bit positions because there are less fields included) and/or defined to have a given field interpreted differently. Thus, each instruction of an ISA is expressed using a given instruction format (and, if defined, in a given one of the instruction templates of that instruction format) and includes fields for specifying the operation and the operands. For example, an ADD instruction has a specific opcode and an instruction format that includes an opcode field to specify that opcode and operand fields to select operands (source1/destination and source2); and an occurrence of this ADD instruction in an instruction stream will have specific contents in the operand fields that select specific operands.
Scientific, financial, auto-vectorized general purpose, RMS (recognition, mining, and synthesis), and visual and multimedia applications (e.g., 2D/3D graphics, image processing, video compression/decompression, voice recognition algorithms and audio manipulation) often require the same operation to be performed on a large number of data items (referred to as “data parallelism”). Single Instruction Multiple Data (SIMD) refers to a type of instruction that causes a processor to perform an operation on multiple data items. SIMD technology is especially suited to processors that can logically divide the bits in a register into a number of fixed-sized data elements, each of which represents a separate value. For example, the bits in a 256-bit register may be specified as a source operand to be operated on as four separate 64-bit packed data elements (quad-word (Q) size data elements), eight separate 32-bit packed data elements (double word (D) size data elements), sixteen separate 16-bit packed data elements (word (W) size data elements), or thirty-two separate 8-bit data elements (byte (B) size data elements). This type of data is referred to as packed data type or vector data type, and operands of this data type are referred to as packed data operands or vector operands. In other words, a packed data item or vector refers to a sequence of packed data elements, and a packed data operand or a vector operand is a source or destination operand of a SIMD instruction (also known as a packed data instruction or a vector instruction).
By way of example, one type of SIMD instruction specifies a single vector operation to be performed on two source vector operands in a vertical fashion to generate a destination vector operand (also referred to as a result vector operand) of the same size, with the same number of data elements, and in the same data element order. The data elements in the source vector operands are referred to as source data elements, while the data elements in the destination vector operand are referred to a destination or result data elements. These source vector operands are of the same size and contain data elements of the same width, and thus they contain the same number of data elements. The source data elements in the same bit positions in the two source vector operands form pairs of data elements (also referred to as corresponding data elements; that is, the data element in data element position 0 of each source operand correspond, the data element in data element position 1 of each source operand correspond, and so on). The operation specified by that SIMD instruction is performed separately on each of these pairs of source data elements to generate a matching number of result data elements, and thus each pair of source data elements has a corresponding result data element. Since the operation is vertical and since the result vector operand is the same size, has the same number of data elements, and the result data elements are stored in the same data element order as the source vector operands, the result data elements are in the same bit positions of the result vector operand as their corresponding pair of source data elements in the source vector operands. In addition to this type of SIMD instruction, there are a variety of other types of SIMD instructions (e.g., that has only one or has more than two source vector operands, that operate in a horizontal fashion, that generates a result vector operand that is of a different size, that has a different size data elements, and/or that has a different data element order). It should be understood that the term destination vector operand (or destination operand) is defined as the direct result of performing the operation specified by an instruction, including the storage of that destination operand at a location (be it a register or at a memory address specified by that instruction) so that it may be accessed as a source operand by another instruction (by specification of that same location by the another instruction).
The SIMD technology, such as that employed by the Intel® Core™ processors having an instruction set including x86, MMX™, Streaming SIMD Extensions (SSE), SSE2, SSE3, SSE4.1, and SSE4.2 instructions, has enabled a significant improvement in application performance. An additional set of SIMD extensions, referred to the Advanced Vector Extensions (AVX) (AVX1 and AVX2) and using the Vector Extensions (VEX) coding scheme, has been, has been released and/or published (e.g., see Intel® 64 and IA-32 Architectures Software Developers Manual, October 2011; and see Intel® Advanced Vector Extensions Programming Reference, June 2011).
Embodiments of the instruction(s) described herein may be embodied in different formats. Additionally, exemplary systems, architectures, and pipelines are detailed below. Embodiments of the instruction(s) may be executed on such systems, architectures, and pipelines, but are not limited to those detailed.
VEX encoding allows instructions to have more than two operands, and allows SIMD vector registers to be longer than 128 bits. The use of a VEX prefix provides for three-operand (or more) syntax. For example, previous two-operand instructions performed operations such as A=A+B, which overwrites a source operand. The use of a VEX prefix enables operands to perform nondestructive operations such as A=B+C.
VEX Prefix (Bytes 0-2) 2102 is encoded in a three-byte form. The first byte is the Format Field 2140 (VEX Byte 0, bits [7:0]), which contains an explicit C4 byte value (the unique value used for distinguishing the C4 instruction format). The second-third bytes (VEX Bytes 1-2) include a number of bit fields providing specific capability. Specifically, REX field 2105 (VEX Byte 1, bits [7-5]) consists of a VEX.R bit field (VEX Byte 1, bit [7]-R), VEX.X bit field (VEX byte 1, bit [6]-X), and VEX.B bit field (VEX byte 1, bit [5]-B). Other fields of the instructions encode the lower three bits of the register indexes as is known in the art (rrr, xxx, and bbb), so that Rrrr, Xxxx, and Bbbb may be formed by adding VEX.R, VEX.X, and VEX.B. Opcode map field 2115 (VEX byte 1, bits [4:0]-mmmmm) includes content to encode an implied leading opcode byte. W Field 2164 (VEX byte 2, bit [7]W)—is represented by the notation VEX.W, and provides different functions depending on the instruction. The role of VEX.vvvv 2120 (VEX Byte 2, bits [6:3]-vvvv) may include the following: 1) VEX.vvvv encodes the first source register operand, specified in inverted (1s complement) form and is valid for instructions with 2 or more source operands; 2) VEX.vvvv encodes the destination register operand, specified in is complement form for certain vector shifts; or 3) VEX.vvvv does not encode any operand, the field is reserved and should contain 1111b. If VEX.L 2168 Size field (VEX byte 2, bit [2]-L)=0, it indicates 128 bit vector; if VEX.L=1, it indicates 256 bit vector. Prefix encoding field 2125 (VEX byte 2, bits [1:0]-pp) provides additional bits for the base operation field.
Real Opcode Field 2130 (Byte 3) is also known as the opcode byte. Part of the opcode is specified in this field. MOD R/M Field 2140 (Byte 4) includes MOD field 2142 (bits [7-6]), Reg field 2144 (bits [5-3]), and R/M field 2146 (bits [2-0]). The role of Reg field 2144 may include the following: encoding either the destination register operand or a source register operand (the rrr of Rrrr), or be treated as an opcode extension and not used to encode any instruction operand. The role of R/M field 2146 may include the following: encoding the instruction operand that references a memory address, or encoding either the destination register operand or a source register operand.
Scale, Index, Base (SIB)—The content of Scale field 2150 (Byte 5) includes SS2152 (bits [7-6]), which is used for memory address generation. The contents of SIB.xxx 2154 (bits [5-3]) and SIB.bbb 2156 (bits [2-0]) have been previously referred to with regard to the register indexes Xxxx and Bbbb. The Displacement Field 2162 and the immediate field (IMM8) 2172 contain address data.
A vector friendly instruction format is an instruction format that is suited for vector instructions (e.g., there are certain fields specific to vector operations). While embodiments are described in which both vector and scalar operations are supported through the vector friendly instruction format, alternative embodiments use only vector operations the vector friendly instruction format.
While embodiments of the invention will be described in which the vector friendly instruction format supports the following: a 64 byte vector operand length (or size) with 32 bit (4 byte) or 64 bit (8 byte) data element widths (or sizes) (and thus, a 64 byte vector consists of either 16 doubleword-size elements or alternatively, 8 quadword-size elements); a 64 byte vector operand length (or size) with 16 bit (2 byte) or 8 bit (1 byte) data element widths (or sizes); a 32 byte vector operand length (or size) with 32 bit (4 byte), 64 bit (8 byte), 16 bit (2 byte), or 8 bit (1 byte) data element widths (or sizes); and a 16 byte vector operand length (or size) with 32 bit (4 byte), 64 bit (8 byte), 16 bit (2 byte), or 8 bit (1 byte) data element widths (or sizes); alternative embodiments may support more, less and/or different vector operand sizes (e.g., 256 byte vector operands) with more, less, or different data element widths (e.g., 128 bit (16 byte) data element widths).
The class A instruction templates in
The generic vector friendly instruction format 2200 includes the following fields listed below in the order illustrated in
Register index field 2244—its content, directly or through address generation, specifies the locations of the source and destination operands, be they in registers or in memory. These include a sufficient number of bits to select N registers from a P×Q (e.g. 32×512, 16×128, 32×1024, 64×1024) register file. While in one embodiment N may be up to three sources and one destination register, alternative embodiments may support more or less sources and destination registers (e.g., may support up to two sources where one of these sources also acts as the destination, may support up to three sources where one of these sources also acts as the destination, may support up to two sources and one destination).
Modifier field 2246—its content distinguishes occurrences of instructions in the generic vector instruction format that specify memory access from those that do not; that is, between no memory access 2205 instruction templates and memory access 2220 instruction templates. Memory access operations read and/or write to the memory hierarchy (in some cases specifying the source and/or destination addresses using values in registers), while non-memory access operations do not (e.g., the source and destinations are registers). While in one embodiment this field also selects between three different ways to perform memory address calculations, alternative embodiments may support more, less, or different ways to perform memory address calculations.
Augmentation operation field 2250—its content distinguishes which one of a variety of different operations to be performed in addition to the base operation. This field is context specific. In one embodiment of the invention, this field is divided into a class field 2268, an alpha field 2252, and a beta field 2254. The augmentation operation field 2250 allows common groups of operations to be performed in a single instruction rather than 2, 3, or 4 instructions. Scale field 2260—its content allows for the scaling of the index field's content for memory address generation (e.g., for address generation that uses 2scale*index+base).
Displacement Field 2262A—its content is used as part of memory address generation (e.g., for address generation that uses 2scale*index+base+displacement). Displacement Factor Field 2262B (note that the juxtaposition of displacement field 2262A directly over displacement factor field 2262B indicates one or the other is used)—its content is used as part of address generation; it specifies a displacement factor that is to be scaled by the size of a memory access (N)—where N is the number of bytes in the memory access (e.g., for address generation that uses 2scale*index+base+scaled displacement). Redundant low-order bits are ignored and hence, the displacement factor field's content is multiplied by the memory operands total size (N) in order to generate the final displacement to be used in calculating an effective address. The value of N is determined by the processor hardware at runtime based on the full opcode field 2274 (described later herein) and the data manipulation field 2254C. The displacement field 2262A and the displacement factor field 2262B are optional in the sense that they are not used for the no memory access 2205 instruction templates and/or different embodiments may implement only one or none of the two.
Data element width field 2264—its content distinguishes which one of a number of data element widths is to be used (in some embodiments for all instructions; in other embodiments for only some of the instructions). This field is optional in the sense that it is not needed if only one data element width is supported and/or data element widths are supported using some aspect of the opcodes.
Write mask field 2270—its content controls, on a per data element position basis, whether that data element position in the destination vector operand reflects the result of the base operation and augmentation operation. Class A instruction templates support merging-writemasking, while class B instruction templates support both merging- and zeroing-writemasking. When merging, vector masks allow any set of elements in the destination to be protected from updates during the execution of any operation (specified by the base operation and the augmentation operation); in other one embodiment, preserving the old value of each element of the destination where the corresponding mask bit has a 0. In contrast, when zeroing vector masks allow any set of elements in the destination to be zeroed during the execution of any operation (specified by the base operation and the augmentation operation); in one embodiment, an element of the destination is set to 0 when the corresponding mask bit has a 0 value. A subset of this functionality is the ability to control the vector length of the operation being performed (that is, the span of elements being modified, from the first to the last one); however, it is not necessary that the elements that are modified be consecutive. Thus, the write mask field 2270 allows for partial vector operations, including loads, stores, arithmetic, logical, etc. While embodiments of the invention are described in which the write mask field's 2270 content selects one of a number of write mask registers that contains the write mask to be used (and thus the write mask field's 2270 content indirectly identifies that masking to be performed), alternative embodiments instead or additional allow the mask write field's 2270 content to directly specify the masking to be performed.
Immediate field 2272—its content allows for the specification of an immediate. This field is optional in the sense that is it not present in an implementation of the generic vector friendly format that does not support immediate and it is not present in instructions that do not use an immediate. Class field 2268—its content distinguishes between different classes of instructions. With reference to
In the case of the non-memory access 2205 instruction templates of class A, the alpha field 2252 is interpreted as an RS field 2252A, whose content distinguishes which one of the different augmentation operation types are to be performed (e.g., round 2252A.1 and data transform 2252A.2 are respectively specified for the no memory access, round type operation 2210 and the no memory access, data transform type operation 2215 instruction templates), while the beta field 2254 distinguishes which of the operations of the specified type is to be performed. In the no memory access 2205 instruction templates, the scale field 2260, the displacement field 2262A, and the displacement scale filed 2262B are not present.
In the no memory access full round control type operation 2210 instruction template, the beta field 2254 is interpreted as a round control field 2254A, whose content(s) provide static rounding. While in the described embodiments of the invention the round control field 2254A includes a suppress all floating point exceptions (SAE) field 2256 and a round operation control field 2258, alternative embodiments may support may encode both these concepts into the same field or only have one or the other of these concepts/fields (e.g., may have only the round operation control field 2258).
SAE field 2256—its content distinguishes whether or not to disable the exception event reporting; when the SAE field's 2256 content indicates suppression is enabled, a given instruction does not report any kind of floating-point exception flag and does not raise any floating point exception handler.
Round operation control field 2258—its content distinguishes which one of a group of rounding operations to perform (e.g., Round-up, Round-down, Round-towards-zero and Round-to-nearest). Thus, the round operation control field 2258 allows for the changing of the rounding mode on a per instruction basis. In one embodiment of the invention where a processor includes a control register for specifying rounding modes, the round operation control field's 2250 content overrides that register value.
In the no memory access data transform type operation 2215 instruction template, the beta field 2254 is interpreted as a data transform field 2254B, whose content distinguishes which one of a number of data transforms is to be performed (e.g., no data transform, swizzle, broadcast).
In the case of a memory access 2220 instruction template of class A, the alpha field 2252 is interpreted as an eviction hint field 2252B, whose content distinguishes which one of the eviction hints is to be used (in
Vector memory instructions perform vector loads from and vector stores to memory, with conversion support. As with regular vector instructions, vector memory instructions transfer data from/to memory in a data element-wise fashion, with the elements that are actually transferred is dictated by the contents of the vector mask that is selected as the write mask.
Temporal data is data likely to be reused soon enough to benefit from caching. This is, however, a hint, and different processors may implement it in different ways, including ignoring the hint entirely. Non-temporal data is data unlikely to be reused soon enough to benefit from caching in the 1st-level cache and should be given priority for eviction. This is, however, a hint, and different processors may implement it in different ways, including ignoring the hint entirely.
In the case of the instruction templates of class B, the alpha field 2252 is interpreted as a write mask control (Z) field 2252C, whose content distinguishes whether the write masking controlled by the write mask field 2270 should be a merging or a zeroing.
In the case of the non-memory access 2205 instruction templates of class B, part of the beta field 2254 is interpreted as an RL field 2257A, whose content distinguishes which one of the different augmentation operation types are to be performed (e.g., round 2257A.1 and vector length (VSIZE) 2257A.2 are respectively specified for the no memory access, write mask control, partial round control type operation 2212 instruction template and the no memory access, write mask control, VSIZE type operation 2217 instruction template), while the rest of the beta field 2254 distinguishes which of the operations of the specified type is to be performed. In the no memory access 2205 instruction templates, the scale field 2260, the displacement field 2262A, and the displacement scale filed 2262B are not present.
In the no memory access, write mask control, partial round control type operation 2210 instruction template, the rest of the beta field 2254 is interpreted as a round operation field 2259A and exception event reporting is disabled (a given instruction does not report any kind of floating-point exception flag and does not raise any floating point exception handler).
Round operation control field 2259A—just as round operation control field 2258, its content distinguishes which one of a group of rounding operations to perform (e.g., Round-up, Round-down, Round-towards-zero and Round-to-nearest). Thus, the round operation control field 2259A allows for the changing of the rounding mode on a per instruction basis. In one embodiment of the invention where a processor includes a control register for specifying rounding modes, the round operation control field's 2250 content overrides that register value.
In the no memory access, write mask control, VSIZE type operation 2217 instruction template, the rest of the beta field 2254 is interpreted as a vector length field 2259B, whose content distinguishes which one of a number of data vector lengths is to be performed on (e.g., 128, 256, or 512 byte).
In the case of a memory access 2220 instruction template of class B, part of the beta field 2254 is interpreted as a broadcast field 2257B, whose content distinguishes whether or not the broadcast type data manipulation operation is to be performed, while the rest of the beta field 2254 is interpreted the vector length field 2259B. The memory access 2220 instruction templates include the scale field 2260, and optionally the displacement field 2262A or the displacement scale field 2262B.
With regard to the generic vector friendly instruction format 2200, a full opcode field 2274 is shown including the format field 2240, the base operation field 2242, and the data element width field 2264. While one embodiment is shown where the full opcode field 2274 includes all of these fields, the full opcode field 2274 includes less than all of these fields in embodiments that do not support all of them. The full opcode field 2274 provides the operation code (opcode).
The augmentation operation field 2250, the data element width field 2264, and the write mask field 2270 allow these features to be specified on a per instruction basis in the generic vector friendly instruction format. The combination of write mask field and data element width field create typed instructions in that they allow the mask to be applied based on different data element widths.
The various instruction templates found within class A and class B are beneficial in different situations. In some embodiments of the invention, different processors or different cores within a processor may support only class A, only class B, or both classes. For instance, a high performance general purpose out-of-order core intended for general-purpose computing may support only class B, a core intended primarily for graphics and/or scientific (throughput) computing may support only class A, and a core intended for both may support both (of course, a core that has some mix of templates and instructions from both classes but not all templates and instructions from both classes is within the purview of the invention). Also, a single processor may include multiple cores, all of which support the same class or in which different cores support different class. For instance, in a processor with separate graphics and general purpose cores, one of the graphics cores intended primarily for graphics and/or scientific computing may support only class A, while one or more of the general purpose cores may be high performance general purpose cores with out of order execution and register renaming intended for general-purpose computing that support only class B. Another processor that does not have a separate graphics core, may include one more general purpose in-order or out-of-order cores that support both class A and class B. Of course, features from one class may also be implemented in the other class in different embodiments of the invention. Programs written in a high level language would be put (e.g., just in time compiled or statically compiled) into an variety of different executable forms, including: 1) a form having only instructions of the class(es) supported by the target processor for execution; or 2) a form having alternative routines written using different combinations of the instructions of all classes and having control flow code that selects the routines to execute based on the instructions supported by the processor which is currently executing the code.
It should be understood that, although embodiments of the invention are described with reference to the specific vector friendly instruction format 2300 in the context of the generic vector friendly instruction format 2200 for illustrative purposes, the invention is not limited to the specific vector friendly instruction format 2300 except where claimed. For example, the generic vector friendly instruction format 2200 contemplates a variety of possible sizes for the various fields, while the specific vector friendly instruction format 2300 is shown as having fields of specific sizes. By way of specific example, while the data element width field 2264 is illustrated as a one bit field in the specific vector friendly instruction format 2300, the invention is not so limited (that is, the generic vector friendly instruction format 2200 contemplates other sizes of the data element width field 2264).
The generic vector friendly instruction format 2200 includes the following fields listed below in the order illustrated in
REX field 2305 (EVEX Byte 1, bits [7-5])—consists of a EVEX.R bit field (EVEX Byte 1, bit [7]-R), EVEX.X bit field (EVEX byte 1, bit [6]-X), and 2257 BEX byte 1, bit [5]-B). The EVEX.R, EVEX.X, and EVEX.B bit fields provide the same functionality as the corresponding VEX bit fields, and are encoded using 1s complement form, i.e. ZMM0 is encoded as 1111B, ZMM15 is encoded as 0000B. Other fields of the instructions encode the lower three bits of the register indexes as is known in the art (rrr, xxx, and bbb), so that Rrrr, Xxxx, and Bbbb may be formed by adding EVEX.R, EVEX.X, and EVEX.B.
REX′ field 2210—this is the first part of the REX′ field 2210 and is the EVEX.R′ bit field (EVEX Byte 1, bit [4]-R′) that is used to encode either the upper 16 or lower 16 of the extended 32 register set. In one embodiment of the invention, this bit, along with others as indicated below, is stored in bit inverted format to distinguish (in the well-known x86 32-bit mode) from the BOUND instruction, whose real opcode byte is 62, but does not accept in the MOD R/M field (described below) the value of 11 in the MOD field; alternative embodiments of the invention do not store this and the other indicated bits below in the inverted format. A value of 1 is used to encode the lower 16 registers. In other words, R′Rrrr is formed by combining EVEX.R′, EVEX.R, and the other RRR from other fields.
Opcode map field 2315 (EVEX byte 1, bits [3:0]-mmmm)—its content encodes an implied leading opcode byte (0F, 0F 38, or 0F 3). Data element width field 2264 (EVEX byte 2, bit [7]-W)—is represented by the notation EVEX.W. EVEX.W is used to define the granularity (size) of the datatype (either 32-bit data elements or 64-bit data elements). EVEX.vvvv 2320 (EVEX Byte 2, bits [6:3]-vvvv)—the role of EVEX.vvvv may include the following: 1) EVEX.vvvv encodes the first source register operand, specified in inverted (1s complement) form and is valid for instructions with 2 or more source operands; 2) EVEX.vvvv encodes the destination register operand, specified in is complement form for certain vector shifts; or 3) EVEX.vvvv does not encode any operand, the field is reserved and should contain 1111b. Thus, EVEX.vvvv field 2320 encodes the 4 low-order bits of the first source register specifier stored in inverted (1s complement) form. Depending on the instruction, an extra different EVEX bit field is used to extend the specifier size to 32 registers. EVEX.U 2268 Class field (EVEX byte 2, bit [2]-U)—If EVEX.U=0, it indicates class A or EVEX.U0; if EVEX.U=1, it indicates class B or EVEX.U1.
Prefix encoding field 2325 (EVEX byte 2, bits [1:0]-pp)—provides additional bits for the base operation field. In addition to providing support for the legacy SSE instructions in the EVEX prefix format, this also has the benefit of compacting the SIMD prefix (rather than requiring a byte to express the SIMD prefix, the EVEX prefix requires only 2 bits). In one embodiment, to support legacy SSE instructions that use a SIMD prefix (66H, F2H, F3H) in both the legacy format and in the EVEX prefix format, these legacy SIMD prefixes are encoded into the SIMD prefix encoding field; and at runtime are expanded into the legacy SIMD prefix prior to being provided to the decoder's PLA (so the PLA can execute both the legacy and EVEX format of these legacy instructions without modification). Although newer instructions could use the EVEX prefix encoding field's content directly as an opcode extension, certain embodiments expand in a similar fashion for consistency but allow for different meanings to be specified by these legacy SIMD prefixes. An alternative embodiment may redesign the PLA to support the 2 bit SIMD prefix encodings, and thus not require the expansion.
Alpha field 2252 (EVEX byte 3, bit [7]-EH; also known as EVEX.EH, EVEX.rs, EVEX.RL, EVEX.write mask control, and EVEX.N; also illustrated with α)—as previously described, this field is context specific. Beta field 2254 (EVEX byte 3, bits [6:4]-SSS, also known as EVEX.s2-0, EVEX.r2-0, EVEX.rr1, EVEX.LL0, EVEX.LLB; also illustrated with βββ)—as previously described, this field is context specific.
REX′ field 2210—this is the remainder of the REX′ field and is the EVEX.V′ bit field (EVEX Byte 3, bit [3]-V′) that may be used to encode either the upper 16 or lower 16 of the extended 32 register set. This bit is stored in bit inverted format. A value of 1 is used to encode the lower 16 registers. In other words, V′VVVV is formed by combining EVEX.V′, EVEX.vvvv.
Write mask field 2270 (EVEX byte 3, bits [2:0]-kkk)—its content specifies the index of a register in the write mask registers as previously described. In one embodiment of the invention, the specific value EVEX.kkk=000 has a special behavior implying no write mask is used for the particular instruction (this may be implemented in a variety of ways including the use of a write mask hardwired to all ones or hardware that bypasses the masking hardware).
Real Opcode Field 2330 (Byte 4) is also known as the opcode byte. Part of the opcode is specified in this field. MOD R/M Field 2340 (Byte 5) includes MOD field 2342, Reg field 2344, and R/M field 2346. As previously described, the MOD field's 2342 content distinguishes between memory access and non-memory access operations. The role of Reg field 2344 can be summarized to two situations: encoding either the destination register operand or a source register operand, or be treated as an opcode extension and not used to encode any instruction operand. The role of R/M field 2346 may include the following: encoding the instruction operand that references a memory address, or encoding either the destination register operand or a source register operand.
Scale, Index, Base (SIB) Byte (Byte 6)—As previously described, the scale field's 2250 content is used for memory address generation. SIB.xxx 2354 and SIB.bbb 2356—the contents of these fields have been previously referred to with regard to the register indexes Xxxx and Bbbb. Displacement field 2262A (Bytes 7-10)—when MOD field 2342 contains 10, bytes 7-10 are the displacement field 2262A, and it works the same as the legacy 32-bit displacement (disp32) and works at byte granularity.
Displacement factor field 2262B (Byte 7)—when MOD field 2342 contains 01, byte 7 is the displacement factor field 2262B. The location of this field is that same as that of the legacy x86 instruction set 8-bit displacement (disp8), which works at byte granularity. Since disp8 is sign extended, it can only address between −128 and 127 bytes offsets; in terms of 64 byte cache lines, disp8 uses 8 bits that can be set to only four really useful values −128, −64, 0, and 64; since a greater range is often needed, disp32 is used; however, disp32 requires 4 bytes. In contrast to disp8 and disp32, the displacement factor field 2262B is a reinterpretation of disp8; when using displacement factor field 2262B, the actual displacement is determined by the content of the displacement factor field multiplied by the size of the memory operand access (N). This type of displacement is referred to as disp8*N. This reduces the average instruction length (a single byte of used for the displacement but with a much greater range). Such compressed displacement is based on the assumption that the effective displacement is multiple of the granularity of the memory access, and hence, the redundant low-order bits of the address offset do not need to be encoded. In other words, the displacement factor field 2262B substitutes the legacy x86 instruction set 8-bit displacement. Thus, the displacement factor field 2262B is encoded the same way as an x86 instruction set 8-bit displacement (so no changes in the ModRM/SIB encoding rules) with the only exception that disp8 is overloaded to disp8*N. In other words, there are no changes in the encoding rules or encoding lengths but only in the interpretation of the displacement value by hardware (which needs to scale the displacement by the size of the memory operand to obtain a byte-wise address offset) Immediate field 2272 operates as previously described.
When U=1, the alpha field 2252 (EVEX byte 3, bit [7]-EH) is interpreted as the write mask control (Z) field 2252C. When U=1 and the MOD field 2342 contains 11 (signifying a no memory access operation), part of the beta field 2254 (EVEX byte 3, bit [4]-S0) is interpreted as the RL field 2257A; when it contains a 1 (round 2257A.1) the rest of the beta field 2254 (EVEX byte 3, bit [6-5]-S2-1) is interpreted as the round operation field 2259A, while when the RL field 2257A contains a 0 (VSIZE 2257.A2) the rest of the beta field 2254 (EVEX byte 3, bit [6-5]-S2-1) is interpreted as the vector length field 2259B (EVEX byte 3, bit [6-5]-L1-0). When U=1 and the MOD field 2342 contains 00, 01, or 10 (signifying a memory access operation), the beta field 2254 (EVEX byte 3, bits [6:4]-SSS) is interpreted as the vector length field 2259B (EVEX byte 3, bit [6-5]-L1-0) and the broadcast field 2257B (EVEX byte 3, bit [4]-B).
In other words, the vector length field 2259B selects between a maximum length and one or more other shorter lengths, where each such shorter length is half the length of the preceding length; and instructions templates without the vector length field 2259B operate on the maximum vector length. Further, in one embodiment, the class B instruction templates of the specific vector friendly instruction format 2300 operate on packed or scalar single/double-precision floating point data and packed or scalar integer data. Scalar operations are operations performed on the lowest order data element position in an zmm/ymm/xmm register; the higher order data element positions are either left the same as they were prior to the instruction or zeroed depending on the embodiment.
Write mask registers 2415—in the embodiment illustrated, there are 8 write mask registers (k0 through k7), each 64 bits in size. In an alternate embodiment, the write mask registers 2415 are 16 bits in size. As previously described, in one embodiment of the invention, the vector mask register k0 cannot be used as a write mask; when the encoding that would normally indicate k0 is used for a write mask, it selects a hardwired write mask of 0xFFFF, effectively disabling write masking for that instruction.
General-purpose registers 2425—in the embodiment illustrated, there are sixteen 64-bit general-purpose registers that are used along with the existing x86 addressing modes to address memory operands. These registers are referenced by the names RAX, RBX, RCX, RDX, RBP, RSI, RDI, RSP, and R8 through R15.
Scalar floating point stack register file (x87 stack) 2445, on which is aliased the MMX packed integer flat register file 2450—in the embodiment illustrated, the x87 stack is an eight-element stack used to perform scalar floating-point operations on 32/64/80-bit floating point data using the x87 instruction set extension; while the MMX registers are used to perform operations on 64-bit packed integer data, as well as to hold operands for some operations performed between the MMX and XMM registers.
Alternative embodiments of the invention may use wider or narrower registers. Additionally, alternative embodiments of the invention may use more, less, or different register files and registers.
Processor cores may be implemented in different ways, for different purposes, and in different processors. For instance, implementations of such cores may include: 1) a general purpose in-order core intended for general-purpose computing; 2) a high performance general purpose out-of-order core intended for general-purpose computing; 3) a special purpose core intended primarily for graphics and/or scientific (throughput) computing Implementations of different processors may include: 1) a CPU including one or more general purpose in-order cores intended for general-purpose computing and/or one or more general purpose out-of-order cores intended for general-purpose computing; and 2) a coprocessor including one or more special purpose cores intended primarily for graphics and/or scientific (throughput). Such different processors lead to different computer system architectures, which may include: 1) the coprocessor on a separate chip from the CPU; 2) the coprocessor on a separate die in the same package as a CPU; 3) the coprocessor on the same die as a CPU (in which case, such a coprocessor is sometimes referred to as special purpose logic, such as integrated graphics and/or scientific (throughput) logic, or as special purpose cores); and 4) a system on a chip that may include on the same die the described CPU (sometimes referred to as the application core(s) or application processor(s)), the above described coprocessor, and additional functionality. Exemplary core architectures are described next, followed by descriptions of processors and computer architectures.
In
The front end unit 2530 includes a branch prediction unit 2532 coupled to an instruction cache unit 2534, which is coupled to an instruction translation lookaside buffer (TLB) 2536, which is coupled to an instruction fetch unit 2538, which is coupled to a decode unit 2540. The decode unit 2540 (or decoder) may decode instructions, and generate as an output one or more micro-operations, micro-code entry points, microinstructions, other instructions, or other control signals, which are decoded from, or which otherwise reflect, or are derived from, the original instructions. The decode unit 2540 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. In one embodiment, the core 2590 includes a microcode ROM or other medium that stores microcode for certain macroinstructions (e.g., in decode unit 2540 or otherwise within the front end unit 2530). The decode unit 2540 is coupled to a rename/allocator unit 2552 in the execution engine unit 2550.
The execution engine unit 2550 includes the rename/allocator unit 2552 coupled to a retirement unit 2554 and a set of one or more scheduler unit(s) 2556. The scheduler unit(s) 2556 represents any number of different schedulers, including reservations stations, central instruction window, etc. The scheduler unit(s) 2556 is coupled to the physical register file(s) unit(s) 2558. Each of the physical register file(s) units 2558 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), etc.
In one embodiment, the physical register file(s) unit 2558 comprises a vector registers unit, a write mask registers unit, and a scalar registers unit. These register units may provide architectural vector registers, vector mask registers, and general purpose registers. The physical register file(s) unit(s) 2558 is overlapped by the retirement unit 2554 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 unit 2554 and the physical register file(s) unit(s) 2558 are coupled to the execution cluster(s) 2560.
The execution cluster(s) 2560 includes a set of one or more execution units 2562 and a set of one or more memory access units 2564. The execution units 2562 may 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 scheduler unit(s) 2556, physical register file(s) unit(s) 2558, and execution cluster(s) 2560 are shown as being possibly plural because certain embodiments create separate pipelines for certain types of data/operations (e.g., a scalar integer pipeline, a scalar floating point/packed integer/packed floating point/vector integer/vector floating point pipeline, and/or a memory access pipeline that each have their own scheduler unit, physical register file(s) unit, and/or execution cluster—and in the case of a separate memory access pipeline, certain embodiments are implemented in which only the execution cluster of this pipeline has the memory access unit(s) 2564). It should also be understood that where separate pipelines are used, one or more of these pipelines may be out-of-order issue/execution and the rest in-order.
The set of memory access units 2564 is coupled to the memory unit 2570, which includes a data TLB unit 2572 coupled to a data cache unit 2574 coupled to a level 2 (L2) cache unit 2576. In one embodiment, the memory access units 2564 may include a load unit, a store address unit, and a store data unit, each of which is coupled to the data TLB unit 2572 in the memory unit 2570. The instruction cache unit 2534 is further coupled to a level 2 (L2) cache unit 2576 in the memory unit 2570. The L2 cache unit 2576 is coupled to one or more other levels of cache and eventually to a main memory.
By way of example, the register renaming, out-of-order issue/execution core architecture may implement the pipeline 2500 as follows: 1) the instruction fetch 2538 performs the fetch and length decoding stages 2502 and 2504; 2) the decode unit 2540 performs the decode stage 2506; 3) the rename/allocator unit 2552 performs the allocation stage 2508 and renaming stage 2510; 4) the scheduler unit(s) 2556 performs the schedule stage 2512; 5) the physical register file(s) unit(s) 2558 and the memory unit 2570 perform the register read/memory read stage 2514; the execution cluster 2560 perform the execute stage 2516; 6) the memory unit 2570 and the physical register file(s) unit(s) 2558 perform the write back/memory write stage 2518; 7) various units may be involved in the exception handling stage 2522; and 8) the retirement unit 2554 and the physical register file(s) unit(s) 2558 perform the commit stage 2524.
The core 2590 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, Calif.; the ARM instruction set (with optional additional extensions such as NEON) of ARM Holdings of Sunnyvale, Calif.), including the instruction(s) described herein. In one embodiment, the core 2590 includes logic to support a packed data instruction set extension (e.g., AVX1, AVX2, and/or some form of the generic vector friendly instruction format (U=0 and/or U=1) previously described), 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).
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 also includes separate instruction and data cache units 2534/2574 and a shared L2 cache unit 2576, 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 local subset of the L2 cache 2604 is part of a global L2 cache that is divided into separate local subsets, one per processor core. Each processor core has a direct access path to its own local subset of the L2 cache 2604. Data read by a processor core is stored in its L2 cache subset 2604 and can be accessed quickly, in parallel with other processor cores accessing their own local L2 cache subsets. Data written by a processor core is stored in its own L2 cache subset 2604 and is flushed from other subsets, if necessary. The ring network ensures coherency for shared data. The ring network is bi-directional to allow agents such as processor cores, L2 caches and other logic blocks to communicate with each other within the chip. Each ring data-path is 1012-bits wide per direction.
Thus, different implementations of the processor 2700 may include: 1) a CPU with the special purpose logic 2708 being integrated graphics and/or scientific (throughput) logic (which may include one or more cores), and the cores 2702A-N being one or more general purpose cores (e.g., general purpose in-order cores, general purpose out-of-order cores, a combination of the two); 2) a coprocessor with the cores 2702A-N being a large number of special purpose cores intended primarily for graphics and/or scientific (throughput); and 3) a coprocessor with the cores 2702A-N being a large number of general purpose in-order cores. Thus, the processor 2700 may be a general-purpose processor, coprocessor or special-purpose processor, such as, for example, a network or communication processor, compression engine, graphics processor, GPGPU (general purpose graphics processing unit), a high-throughput many integrated core (MIC) coprocessor (including 30 or more cores), embedded processor, or the like. The processor may be implemented on one or more chips. The processor 2700 may be a part of and/or may be implemented on one or more substrates using any of a number of process technologies, such as, for example, BiCMOS, CMOS, or NMOS.
The memory hierarchy includes one or more levels of cache within the cores, a set or one or more shared cache units 2706, and external memory (not shown) coupled to the set of integrated memory controller units 2714. The set of shared cache units 2706 may include one or more mid-level caches, such as level 2 (L2), level 3 (L3), level 4 (L4), or other levels of cache, a last level cache (LLC), and/or combinations thereof. While in one embodiment a ring based interconnect unit 2712 interconnects the integrated graphics logic 2708, the set of shared cache units 2706, and the system agent unit 2710/integrated memory controller unit(s) 2714, alternative embodiments may use any number of well-known techniques for interconnecting such units. In one embodiment, coherency is maintained between one or more cache units 2706 and cores 2702-A-N.
In some embodiments, one or more of the cores 2702A-N are capable of multi-threading. The system agent 2710 includes those components coordinating and operating cores 2702A-N. The system agent unit 2710 may include for example a power control unit (PCU) and a display unit. The PCU may be or include logic and components needed for regulating the power state of the cores 2702A-N and the integrated graphics logic 2708. The display unit is for driving one or more externally connected displays.
The cores 2702A-N may be homogenous or heterogeneous in terms of architecture instruction set; that is, two or more of the cores 2702A-N may be capable of execution the same instruction set, while others may be capable of executing only a subset of that instruction set or a different instruction set.
Referring now to
The optional nature of additional processors 2815 is denoted in
The memory 2840 may be, for example, dynamic random access memory (DRAM), phase change memory (PCM), or a combination of the two. For at least one embodiment, the controller hub 2820 communicates with the processor(s) 2810, 2815 via a multi-drop bus, such as a frontside bus (FSB), point-to-point interface such as QuickPath Interconnect (QPI), or similar connection 2895.
In one embodiment, the coprocessor 2845 is a special-purpose processor, such as, for example, a high-throughput MIC processor, a network or communication processor, compression engine, graphics processor, GPGPU, embedded processor, or the like. In one embodiment, controller hub 2820 may include an integrated graphics accelerator.
There can be a variety of differences between the physical resources 2810, 2815 in terms of a spectrum of metrics of merit including architectural, microarchitectural, thermal, power consumption characteristics, and the like.
In one embodiment, the processor 2810 executes instructions that control data processing operations of a general type. Embedded within the instructions may be coprocessor instructions. The processor 2810 recognizes these coprocessor instructions as being of a type that should be executed by the attached coprocessor 2845. Accordingly, the processor 2810 issues these coprocessor instructions (or control signals representing coprocessor instructions) on a coprocessor bus or other interconnect, to coprocessor 2845. Coprocessor(s) 2845 accept and execute the received coprocessor instructions.
Referring now to
Processors 2970 and 2980 are shown including integrated memory controller (IMC) units 2972 and 2982, respectively. Processor 2970 also includes as part of its bus controller units point-to-point (P-P) interfaces 2976 and 2978; similarly, second processor 2980 includes P-P interfaces 2986 and 2988. Processors 2970, 2980 may exchange information via a point-to-point (P-P) interface 2950 using P-P interface circuits 2978, 2988. As shown in
Processors 2970, 2980 may each exchange information with a chipset 2990 via individual P-P interfaces 2952, 2954 using point to point interface circuits 2976, 2994, 2986, 2998. Chipset 2990 may optionally exchange information with the coprocessor 2938 via a high-performance interface 2939. In one embodiment, the coprocessor 2938 is a special-purpose processor, such as, for example, a high-throughput MIC processor, a network or communication processor, compression engine, graphics processor, GPGPU, embedded processor, or the like.
A shared cache (not shown) may be included in either processor or outside of both processors, yet connected with the processors via P-P interconnect, such that either or both processors' local cache information may be stored in the shared cache if a processor is placed into a low power mode. Chipset 2990 may be coupled to a first bus 2916 via an interface 2996. In one embodiment, first bus 2916 may be a Peripheral Component Interconnect (PCI) bus, or a bus such as a PCI Express bus or another third generation I/O interconnect bus, although the scope of the present invention is not so limited.
As shown in
Referring now to
Referring now to
Embodiments of the mechanisms disclosed herein may be implemented in hardware, software, firmware, or a combination of such implementation approaches. Embodiments of the invention may be implemented as computer programs or program code executing on programmable systems comprising at least one processor, a storage system (including volatile and non-volatile memory and/or storage elements), at least one input device, and at least one output device.
Program code, such as code 2930 illustrated in
The program code may be implemented in a high level procedural or object oriented programming language to communicate with a processing system. The program code may also be implemented in assembly or machine language, if desired. In fact, the mechanisms described herein are not limited in scope to any particular programming language. In any case, the language may be a compiled or interpreted language.
One or more aspects of at least one embodiment may be implemented by representative instructions stored on a machine-readable medium which represents various logic within the processor, which when read by a machine causes the machine to fabricate logic to perform the techniques described herein. Such representations, known as “IP cores” may be stored on a tangible, machine readable medium and supplied to various customers or manufacturing facilities to load into the fabrication machines that actually make the logic or processor.
Such machine-readable storage media may include, without limitation, non-transitory, tangible arrangements of articles manufactured or formed by a machine or device, including storage media such as hard disks, any other type of disk including floppy disks, optical disks, compact disk read-only memories (CD-ROMs), compact disk rewritable's (CD-RWs), and magneto-optical disks, semiconductor devices such as read-only memories (ROMs), random access memories (RAMs) such as dynamic random access memories (DRAMs), static random access memories (SRAMs), erasable programmable read-only memories (EPROMs), flash memories, electrically erasable programmable read-only memories (EEPROMs), phase change memory (PCM), magnetic or optical cards, or any other type of media suitable for storing electronic instructions.
Accordingly, embodiments of the invention also include non-transitory, tangible machine-readable media containing instructions or containing design data, such as Hardware Description Language (HDL), which defines structures, circuits, apparatuses, processors and/or system features described herein. Such embodiments may also be referred to as program products.
In some cases, an instruction converter may be used to convert an instruction from a source instruction set to a target instruction set. For example, the instruction converter may translate (e.g., using static binary translation, dynamic binary translation including dynamic compilation), morph, emulate, or otherwise convert an instruction to one or more other instructions to be processed by the core. The instruction converter may be implemented in software, hardware, firmware, or a combination thereof. The instruction converter may be on processor, off processor, or part on and part off processor.
According to one embodiment, a processor includes an instruction decoder to decode a first instruction to gather data elements from memory, the first instruction having a first operand specifying a first storage location and a second operand specifying a first memory address storing a plurality of data elements. The processor further includes an execution unit coupled to the instruction decoder, in response to the first instruction, to read contiguous a first and a second of the data elements from a memory location based on the first memory address indicated by the second operand, and to store the first data element in a first entry of the first storage location and a second data element in a second entry of a second storage location corresponding to the first entry of the first storage location. In one embodiment, the first instruction further comprises a third operand specifying the second storage location. In one embodiment, the instruction decoder further decodes a second instruction having a third operand specifying the second storage location, and a fourth operand specifying a second memory address, the second memory address being offset from the first memory address by the size of a single data element. According to one aspect of the invention, the first instruction further comprises a prefix indicating to the instruction decoder and execution unit that the second instruction follows. In another embodiment, the execution unit predicts the existence of the second instruction. In one embodiment, the first entry of the first storage location is not contiguous to the second entry of the second storage location, and wherein the second storage location is specified by the first operand. According to one embodiment, the first data element is stored in a third entry of a third storage location prior to being stored in the first entry of the first storage location, and the second data element is stored in a fourth entry of a fourth storage location prior to being stored in the second entry of the second storage location.
Some portions of the preceding detailed descriptions have been presented in terms of algorithms and symbolic representations of operations on data bits within a computer memory. These algorithmic descriptions and representations are the ways used by those skilled in the data processing arts to most effectively convey the substance of their work to others skilled in the art. An algorithm is here, and generally, conceived to be a self-consistent sequence of operations leading to a desired result. The operations are those requiring physical manipulations of physical quantities.
It should be borne in mind, however, that all of these and similar terms are to be associated with the appropriate physical quantities and are merely convenient labels applied to these quantities. Unless specifically stated otherwise as apparent from the above discussion, it is appreciated that throughout the description, discussions utilizing terms such as those set forth in the claims below, refer to the action and processes of a computer system, or similar electronic computing device, that manipulates and transforms data represented as physical (electronic) quantities within the computer system's registers and memories into other data similarly represented as physical quantities within the computer system memories or registers or other such information storage, transmission or display devices.
The techniques shown in the figures can be implemented using code and data stored and executed on one or more electronic devices. Such electronic devices store and communicate (internally and/or with other electronic devices over a network) code and data using computer-readable media, such as non-transitory computer-readable storage media (e.g., magnetic disks; optical disks; random access memory; read only memory; flash memory devices; phase-change memory) and transitory computer-readable transmission media (e.g., electrical, optical, acoustical or other form of propagated signals—such as carrier waves, infrared signals, digital signals).
The processes or methods depicted in the preceding figures may be performed by processing logic that comprises hardware (e.g. circuitry, dedicated logic, etc.), firmware, software (e.g., embodied on a non-transitory computer readable medium), or a combination of both. Although the processes or methods are described above in terms of some sequential operations, it should be appreciated that some of the operations described may be performed in a different order. Moreover, some operations may be performed in parallel rather than sequentially.
In the foregoing specification, embodiments of the invention have been described with reference to specific embodiments thereof. It will be evident that various modifications may be made thereto without departing from the broader spirit and scope of the invention as set forth in the following claims. The specification and drawings are, accordingly, to be regarded in an illustrative sense rather than a restrictive sense.
The present application is a continuation of U.S. patent application Ser. No. 13/997,784, filed on Jun. 25, 2013, entitled “COALESCING ADJACENT GATHER/SCATTER OPERATIONS”, which is hereby incorporated herein by this reference in its entirety and for all purposes. U.S. patent application Ser. No. 13/997,784 is itself a U.S. National Phase Application under 35 U.S.C. §371 of International Application No. PCT/US2012/071688, filed Dec. 26, 2012, entitled COALESCING ADJACENT GATHER/SCATTER OPERATIONS.
| Number | Name | Date | Kind |
|---|---|---|---|
| 5734874 | Van Hook et al. | Mar 1998 | A |
| 9348601 | Forsyth et al. | May 2016 | B2 |
| 20020199086 | Wilson | Dec 2002 | A1 |
| 20030023960 | Khan et al. | Jan 2003 | A1 |
| 20090172364 | Sprangle et al. | Jul 2009 | A1 |
| 20090198885 | Manoj | Aug 2009 | A1 |
| 20100030756 | Ahmed et al. | Feb 2010 | A1 |
| 20100169281 | Atluri et al. | Jul 2010 | A1 |
| 20100191712 | Wolman et al. | Jul 2010 | A1 |
| 20110153983 | Hughes et al. | Jun 2011 | A1 |
| 20110153993 | Gopal | Jun 2011 | A1 |
| 20110264863 | Sprangle et al. | Oct 2011 | A1 |
| 20120137074 | Kim et al. | May 2012 | A1 |
| 20120144089 | Hall et al. | Jun 2012 | A1 |
| 20120166761 | Hughes et al. | Jun 2012 | A1 |
| 20140201498 | Ould-Ahmed-Vall et al. | Jul 2014 | A1 |
| Number | Date | Country |
|---|---|---|
| 10-1099467 | Dec 2011 | KR |
| 2014105011 | Jul 2014 | WO |
| Entry |
|---|
| Final Office Action received for U.S. Appl. No. 13/997,784, mailed on Jul. 8, 2015, 14 pages. |
| Non-Final Office Action received for U.S. Appl. No. 13/997,784, mailed on Mar. 4, 2015, 13 pages. |
| Notice of Allowance received for U.S. Appl. No. 13/997,784, mailed on Jan. 21, 2016, 7 pages. |
| International Preliminary Report on Patentability and Written Opinion Received for PCT Patent Application No. PCT/US2012/071688, mailed on Jul. 9, 2015, 6 pages. |
| International Search Report and Written Opinion received for PCT Patent Application No. PCT/US2012/071688, mailed on Jun. 27, 2013, 9 pages. |
| Office Action received for Korean Patent Application No. 10-2015-7011767, mailed on Feb. 18, 2016, 18 pages of English Translation and 17 pages of Korean Office Action. |
| Non-Final Office Action received for U.S. Appl. No. 14/975,222, mailed on Mar. 10, 2016, 5 pages. |
| Non-Final Office Action Received for U.S. Appl. No. 14/975,327, mailed on May 9, 2016, 11 pages. |
| Non-Final Office Action received for U.S. Appl. No. 14/976,231, mailed on Apr. 21, 2016, 12 pages. |
| ARM, “ARM® Architecture Reference Manual ARM® v7-A and ARM® v7-R edition”, Apr. 5, 2007, pp. A3-A9, A3-A42, A4-A27, A7-A28, A7-A29, A8-A101 to A8-602-A8-625 and A8-768-A8-783. |
| Non Final Office Action received for U.S. Appl. No. 14/976,216, mailed on Jun. 3, 2016, 11 pages. |
| Non Final Office Action received for U.S. Appl. No. 14/976,220, mailed on Jun. 16, 2016, 9 pages. |
| Non Final Office Action received for U.S. Appl. No. 14/975,292, mailed on May 19, 2016, 9 pages. |
| Non Final Office Action received for U.S. Appl. No. 14/976,228, mailed on Jul. 1, 2016, 10 pages. |
| Notice of Allowance received for U.S. Appl. No. 14/975,222, mailed on Sep. 21, 2016, 10 pages. |
| Number | Date | Country | |
|---|---|---|---|
| 20160103788 A1 | Apr 2016 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 13997784 | US | |
| Child | 14975323 | US |
