AVX-512
AVX-512 are 512-bit extensions to the 256-bit Advanced Vector Extensions SIMD instructions for x86 instruction set architecture proposed by Intel in July 2013, and first implemented in the 2016 Intel Xeon Phi x200, and then later in a number of AMD and other Intel CPUs. AVX-512 consists of multiple extensions that may be implemented independently. This policy is a departure from the historical requirement of implementing the entire instruction block. Only the core extension AVX-512F is required by all AVX-512 implementations.
Besides widening most 256-bit instructions, the extensions introduce various new operations, such as new data conversions, scatter operations, and permutations. The number of AVX registers is increased from 16 to 32, and eight new "mask registers" are added, which allow for variable selection and blending of the results of instructions. In CPUs with the vector length extension—included in most AVX-512-capable processors —these instructions may also be used on the 128-bit and 256-bit vector sizes.
AVX-512 is not the first 512-bit SIMD instruction set that Intel has introduced in processors: the earlier 512-bit SIMD instructions used in the first generation Xeon Phi coprocessors, derived from Intel's Larrabee project, are similar but not binary compatible and only partially source compatible.
The successor to AVX-512 is AVX10, announced in July 2023. AVX10 simplifies detection of supported instructions by introducing a version of the instruction set, where each subsequent version includes all instructions from the previous one. In the initial revisions of the AVX10 specification, the support for 512-bit vectors was made optional, which would allow Intel to support it in their E-cores. In later revisions, Intel made 512-bit vectors mandatory, with the intention to support 512-bit vectors both in P- and E-cores. The initial version 1 of AVX10 does not add new instructions compared to AVX-512, and for processors supporting 512-bit vectors it is equivalent to AVX-512. Later AVX10 versions will introduce new features.
Instruction set
The AVX-512 instruction set consists of several separate sets each having their own unique CPUID feature bit. However, they are typically grouped by the processor generation that implements them.F, CD, ER, PF: introduced with Xeon Phi x200 (Knights Landing) and Xeon Scalable, with the last two being specific to Knights Landing & Knights Mill.AVX-512 Foundation – expands most 32-bit and 64-bit based AVX instructions with the EVEX coding scheme to support 512-bit registers, operation masks, parameter broadcasting, and embedded rounding and exception control, implemented by Knights Landing and Skylake XeonAVX-512 Conflict Detection Instructions – efficient conflict detection to allow more loops to be vectorized, implemented by Knights Landing and Skylake XAVX-512 Exponential and Reciprocal Instructions – exponential and reciprocal operations designed to help implement transcendental operations, implemented by Knights LandingAVX-512 Prefetch Instructions – new prefetch capabilities, implemented by Knights Landing
4VNNIW, 4FMAPS: introduced with and specific to Knights Mill.AVX-512 Vector Neural Network Instructions Word variable precision – vector instructions for deep learning, enhanced word, variable precision.AVX-512 Fused Multiply Accumulation Packed Single precision – vector instructions for deep learning, floating point, single precision.
VL, DQ, BW: introduced with Skylake-X/SP and Cannon Lake.AVX-512 Vector Length Extensions – extends most AVX-512 operations to also operate on XMM and YMM registersAVX-512 Doubleword and Quadword Instructions – adds new 32-bit and 64-bit AVX-512 instructionsAVX-512 Byte and Word Instructions – extends AVX-512 to cover 8-bit and 16-bit integer operations
IFMA, VBMI: introduced with Cannon Lake.AVX-512 Integer Fused Multiply Add – fused multiply add of integers using 52-bit precision.AVX-512 Vector Bit Manipulation Instructions adds vector byte permutation instructions which were not present in AVX-512BW.
VNNI: introduced with Cascade Lake.AVX-512 Vector Neural Network Instructions – vector instructions for deep learning.
VPOPCNTDQ: Vector population count instruction. Introduced with Knights Mill and Ice Lake.
VBMI2, BITALG: introduced with Ice Lake.AVX-512 Vector Bit Manipulation Instructions 2 – byte/word load, store and concatenation with shift.AVX-512 Bit Algorithms – byte/word bit manipulation instructions expanding VPOPCNTDQ.
VP2INTERSECT: introduced with Tiger Lake.AVX-512 Vector Pair Intersection to a Pair of Mask Registers .
GFNI, VPCLMULQDQ, VAES: introduced with Ice Lake.
- These are not AVX-512 features per se. Together with AVX-512, they enable EVEX encoded versions of GFNI, PCLMULQDQ and AES instructions.
Encoding and features
The VEX prefix used by AVX and AVX2, while flexible, did not leave enough room for the features Intel wanted to add to AVX-512. This has led them to define a new prefix called EVEX.Compared to VEX, EVEX adds the following benefits:
- Expanded register encoding allowing 32 512-bit registers.
- Adds 8 new opmask registers for masking most AVX-512 instructions.
- Adds a new scalar memory mode that automatically performs a broadcast.
- Adds room for explicit rounding control in each instruction.
- Adds a new compressed displacement memory addressing mode.
SIMD modes
The AVX-512 instructions are designed to mix with 128/256-bit AVX/AVX2 instructions without a performance penalty. However, AVX-512VL extensions allows the use of AVX-512 instructions on 128/256-bit registers XMM/YMM, so most SSE and AVX/AVX2 instructions have new AVX-512 versions encoded with the EVEX prefix which allow access to new features such as opmask and additional registers. Unlike AVX-256, the new instructions do not have new mnemonics but share namespace with AVX, making the distinction between VEX and EVEX encoded versions of an instruction ambiguous in the source code. Since AVX-512F only works on 32- and 64-bit values, SSE and AVX/AVX2 instructions that operate on bytes or words are available only with the AVX-512BW extension.| Name | Extension sets | Registers | Types |
| Legacy SSE | SSE–SSE4.2 | xmm0–xmm15 | single floats from SSE2: bytes, words, doublewords, quadwords and double floats |
| AVX-128 | AVX, AVX2 | xmm0–xmm15 | bytes, words, doublewords, quadwords, single floats and double floats |
| AVX-256 | AVX, AVX2 | ymm0–ymm15 | single float and double float from AVX2: bytes, words, doublewords, quadwords |
| AVX-128 | AVX-512VL | xmm0–xmm31 | doublewords, quadwords, single float and double float with AVX512BW: bytes and words. with AVX512-FP16: half float |
| AVX-256 AVX10/256 | AVX-512VL | ymm0–ymm31 | doublewords, quadwords, single float and double float with AVX512BW: bytes and words. with AVX512-FP16: half float |
| AVX10/512 | AVX-512F | doublewords, quadwords, single float and double float with AVX512BW: bytes and words with AVX512-FP16: half float |
Extended registers
The width of the SIMD register file is increased from 256 bits to 512 bits, and expanded from 16 to a total of 32 registers ZMM0–ZMM31. These registers can be addressed as 256 bit YMM registers from AVX extensions and 128-bit XMM registers from Streaming SIMD Extensions, and legacy AVX and SSE instructions can be extended to operate on the 16 additional registers XMM16-XMM31 and YMM16-YMM31 when using EVEX encoded form.Opmask registers
AVX-512 vector instructions may indicate an opmask register to control which values are written to the destination. The instruction encoding supports 0-7 for this field; however, only opmask registers k1-k7 can be used as the mask corresponding to the value 1-7, whereas the value 0 is reserved for indicating no opmask register is used; that is, a hardcoded constant is used to indicate unmasked operations. The special opmask register 'k0' is still a functioning, valid register, it can be used in opmask register manipulation instructions or used as the destination opmask register. A flag controls the opmask behavior, which can either be "zero", which zeros everything not selected by the mask, or "merge", which leaves everything not selected untouched. The merge behavior is identical to the blend instructions.The opmask registers are normally 16 bits wide, but can be up to 64 bits with the AVX-512BW extension. How many of the bits are actually used, though, depends on the vector type of the instructions masked. For the 32-bit single float or double words, 16 bits are used to mask the 16 elements in a 512-bit register. For double float and quad words, at most 8 mask bits are used.
The opmask register is the reason why several bitwise instructions which naturally have no element widths had them added in AVX-512. For instance, bitwise AND, OR or 128-bit shuffle now exist in both double-word and quad-word variants with the only difference being in the final masking.
New opmask instructions
The opmask registers have a new mini extension of instructions operating directly on them. Unlike the rest of the AVX-512 instructions, these instructions are all VEX encoded. The initial opmask instructions are all 16-bit versions. With AVX-512DQ 8-bit versions were added to better match the needs of masking 8 64-bit values, and with AVX-512BW 32-bit and 64-bit versions were added so they can mask up to 64 8-bit values. The instructions KORTEST and KTEST can be used to set the x86 flags based on mask registers, so that they may be used together with non-SIMD x86 branch and conditional instructions.| Instruction | Extension set | Description |
KAND | F | Bitwise logical AND Masks |
KANDN | F | Bitwise logical AND NOT Masks |
KMOV | F | Move from and to Mask Registers or General Purpose Registers |
KUNPCK | F | Unpack for Mask Registers |
KNOT | F | NOT Mask Register |
KOR | F | Bitwise logical OR Masks |
KORTEST | F | OR Masks And Set Flags |
KSHIFTL | F | Shift Left Mask Registers |
KSHIFTR | F | Shift Right Mask Registers |
KXNOR | F | Bitwise logical XNOR Masks |
KXOR | F | Bitwise logical XOR Masks |
KADD | BW/DQ | Add Two Masks |
KTEST | BW/DQ | Bitwise comparison and set flags |
New instructions in AVX-512 foundation
Many AVX-512 instructions are simply EVEX versions of old SSE or AVX instructions. There are, however, several new instructions, and old instructions that have been replaced with new AVX-512 versions. The new or heavily reworked instructions are listed below. These foundation instructions also include the extensions from AVX-512VL and AVX-512BW since those extensions merely add new versions of these instructions instead of new instructions.Blend using mask
There are no EVEX-prefixed versions of the blend instructions from SSE4; instead, AVX-512 has a new set of blending instructions using mask registers as selectors. Together with the general compare into mask instructions below, these may be used to implement generic ternary operations or cmov, similar to XOP's VPCMOV.Since blending is an integral part of the EVEX encoding, these instructions may also be considered basic move instructions. Using the zeroing blend mode, they can also be used as masking instructions.
| Instruction | Extension set | Description |
VBLENDMPD | F | Blend float64 vectors using opmask control |
VBLENDMPS | F | Blend float32 vectors using opmask control |
VPBLENDMD | F | Blend int32 vectors using opmask control |
VPBLENDMQ | F | Blend int64 vectors using opmask control |
VPBLENDMB | BW | Blend byte integer vectors using opmask control |
VPBLENDMW | BW | Blend word integer vectors using opmask control |
Compare into mask
AVX-512F has four new compare instructions. Like their XOP counterparts they use the immediate field to select between 8 different comparisons. Unlike their XOP inspiration, however, they save the result to a mask register and initially only support doubleword and quadword comparisons. The AVX-512BW extension provides the byte and word versions. Note that two mask registers may be specified for the instructions, one to write to and one to declare regular masking.| Imme- diate | Compa- rison | Description |
| 0 | EQ | Equal |
| 1 | LT | Less than |
| 2 | LE | Less than or equal |
| 3 | FALSE | Set to zero |
| 4 | NEQ | Not equal |
| 5 | NLT | Greater than or equal |
| 6 | NLE | Greater than |
| 7 | TRUE | Set to one |
| Instruction | Extension set | Description |
VPCMPD, VPCMPUD | F | Compare signed/unsigned doublewords into mask |
VPCMPQ, VPCMPUQ | F | Compare signed/unsigned quadwords into mask |
VPCMPB, VPCMPUB | BW | Compare signed/unsigned bytes into mask |
VPCMPW, VPCMPUW | BW | Compare signed/unsigned words into mask |
Logical set mask
The final way to set masks is using Logical Set Mask. These instructions perform either AND or NAND, and then set the destination opmask based on the result values being zero or non-zero. Note that like the comparison instructions, these take two opmask registers, one as destination and one a regular opmask.| Instruction | Extension set | Description |
VPTESTMD, VPTESTMQ | F | Logical AND and set mask for 32 or 64 bit integers. |
VPTESTNMD, VPTESTNMQ | F | Logical NAND and set mask for 32 or 64 bit integers. |
VPTESTMB, VPTESTMW | BW | Logical AND and set mask for 8 or 16 bit integers. |
VPTESTNMB, VPTESTNMW | BW | Logical NAND and set mask for 8 or 16 bit integers. |
Compress and expand
The compress and expand instructions match the APL operations of the same name. They use the opmask in a slightly different way from other AVX-512 instructions. Compress only saves the values marked in the mask, but saves them compacted by skipping and not reserving space for unmarked values. Expand operates in the opposite way, by loading as many values as indicated in the mask and then spreading them to the selected positions.| Instruction | Description |
VCOMPRESSPD, VCOMPRESSPS | Store sparse packed double/single-precision floating-point values into dense memory |
VPCOMPRESSD, VPCOMPRESSQ | Store sparse packed doubleword/quadword integer values into dense memory/register |
VEXPANDPD, VEXPANDPS | Load sparse packed double/single-precision floating-point values from dense memory |
VPEXPANDD, VPEXPANDQ | Load sparse packed doubleword/quadword integer values from dense memory/register |
Permute
A new set of permute instructions have been added for full two input permutations. They all take three arguments, two source registers and one index; the result is output by either overwriting the first source register or the index register. AVX-512BW extends the instructions to also include 16-bit versions, and the AVX-512_VBMI extension defines the byte versions of the instructions.| Instruction | Extension set | Description |
VPERMB | VBMI | Permute packed bytes elements. |
VPERMW | BW | Permute packed words elements. |
VPERMT2B | VBMI | Full byte permute overwriting first source. |
VPERMT2W | BW | Full word permute overwriting first source. |
VPERMI2PD, VPERMI2PS | F | Full single/double floating-point permute overwriting the index. |
VPERMI2D, VPERMI2Q | F | Full doubleword/quadword permute overwriting the index. |
VPERMI2B | VBMI | Full byte permute overwriting the index. |
VPERMI2W | BW | Full word permute overwriting the index. |
VPERMT2PS, VPERMT2PD | F | Full single/double floating-point permute overwriting first source. |
VPERMT2D, VPERMT2Q | F | Full doubleword/quadword permute overwriting first source. |
VSHUFF32x4, VSHUFF64x2,VSHUFI32x4, VSHUFI64x2 | F | Shuffle four packed 128-bit lines. |
VPMULTISHIFTQB | VBMI | Select packed unaligned bytes from quadword sources. |
Bitwise ternary logic
Two new instructions added can logically implement all possible bitwise operations between three inputs. They take three registers as input and an 8-bit immediate field. Each bit in the output is generated using a lookup of the three corresponding bits in the inputs to select one of the 8 positions in the 8-bit immediate. Since only 8 combinations are possible using three bits, this allow all possible 3 input bitwise operations to be performed.These are the only bitwise vector instructions in AVX-512F; EVEX versions of the two source SSE and AVX bitwise vector instructions AND, ANDN, OR and XOR were added in AVX-512DQ.
The difference in the doubleword and quadword versions is only the application of the opmask.
| Instruction | Description |
VPTERNLOGD, VPTERNLOGQ | Bitwise Ternary Logic |
Conversions
A number of conversion or move instructions were added; these complete the set of conversion instructions available from SSE2.| Instruction | Extension set | Description |
VPMOVQD, VPMOVSQD, VPMOVUSQD,VPMOVQW, VPMOVSQW, VPMOVUSQW,VPMOVQB, VPMOVSQB, VPMOVUSQB,VPMOVDW, VPMOVSDW, VPMOVUSDW,VPMOVDB, VPMOVSDB, VPMOVUSDB | F | Down convert quadword or doubleword to doubleword, word or byte; unsaturated, saturated or saturated unsigned. The reverse of the sign/zero extend instructions from SSE4.1. |
VPMOVWB, VPMOVSWB, VPMOVUSWB | BW | Down convert word to byte; unsaturated, saturated or saturated unsigned. |
VCVTPS2UDQ, VCVTPD2UDQ,VCVTTPS2UDQ, VCVTTPD2UDQ | F | Convert with or without truncation, packed single or double-precision floating point to packed unsigned doubleword integers. |
VCVTSS2USI, VCVTSD2USI,VCVTTSS2USI, VCVTTSD2USI | F | Convert with or without truncation, scalar single or double-precision floating point to unsigned doubleword integer. |
VCVTPS2QQ, VCVTPD2QQ,VCVTPS2UQQ, VCVTPD2UQQ,VCVTTPS2QQ, VCVTTPD2QQ,VCVTTPS2UQQ, VCVTTPD2UQQ | DQ | Convert with or without truncation, packed single or double-precision floating point to packed signed or unsigned quadword integers. |
VCVTUDQ2PS, VCVTUDQ2PD | F | Convert packed unsigned doubleword integers to packed single or double-precision floating point. |
VCVTUSI2PS, VCVTUSI2PD | F | Convert scalar unsigned doubleword integers to single or double-precision floating point. |
VCVTUSI2SD, VCVTUSI2SS | F | Convert scalar unsigned integers to single or double-precision floating point. |
VCVTUQQ2PS, VCVTUQQ2PD | DQ | Convert packed unsigned quadword integers to packed single or double-precision floating point. |
VCVTQQ2PD, VCVTQQ2PS | F | Convert packed quadword integers to packed single or double-precision floating point. |
Floating-point decomposition
Among the unique new features in AVX-512F are instructions to decompose floating-point values and handle special floating-point values. Since these methods are completely new, they also exist in scalar versions.| Instruction | Description |
VGETEXPPD, VGETEXPPS | Convert exponents of packed fp values into fp values |
VGETEXPSD, VGETEXPSS | Convert exponent of scalar fp value into fp value |
VGETMANTPD, VGETMANTPS | Extract vector of normalized mantissas from float32/float64 vector |
VGETMANTSD, VGETMANTSS | Extract float32/float64 of normalized mantissa from float32/float64 scalar |
VFIXUPIMMPD, VFIXUPIMMPS | Fix up special packed float32/float64 values |
VFIXUPIMMSD, VFIXUPIMMSS | Fix up special scalar float32/float64 value |
Floating-point arithmetic
This is the second set of new floating-point methods, which includes new scaling and approximate calculation of reciprocal, and reciprocal of square root. The approximate reciprocal instructions guarantee to have at most a relative error of 2−14.| Instruction | Description |
VRCP14PD, VRCP14PS | Compute approximate reciprocals of packed float32/float64 values |
VRCP14SD, VRCP14SS | Compute approximate reciprocals of scalar float32/float64 value |
VRNDSCALEPS, VRNDSCALEPD | Round packed float32/float64 values to include a given number of fraction bits |
VRNDSCALESS, VRNDSCALESD | Round scalar float32/float64 value to include a given number of fraction bits |
VRSQRT14PD, VRSQRT14PS | Compute approximate reciprocals of square roots of packed float32/float64 values |
VRSQRT14SD, VRSQRT14SS | Compute approximate reciprocal of square root of scalar float32/float64 value |
VSCALEFPS, VSCALEFPD | Scale packed float32/float64 values with float32/float64 values |
VSCALEFSS, VSCALEFSD | Scale scalar float32/float64 value with float32/float64 value |
New instructions by sets
Conflict detection
The instructions in AVX-512 conflict detection are designed to help efficiently calculate conflict-free subsets of elements in loops that normally could not be safely vectorized.| Instruction | Name | Description |
VPCONFLICTD,VPCONFLICTQ | Detect conflicts within vector of packed double- or quadwords values | Compares each element in the first source, to all elements on same or earlier places in the second source and forms a bit vector of the results |
VPLZCNTD,VPLZCNTQ | Count the number of leading zero bits for packed double- or quadword values | Vectorized LZCNT instruction |
VPBROADCASTMB2Q,VPBROADCASTMW2D | Broadcast mask to vector register | Either 8-bit mask to quadword vector, or 16-bit mask to doubleword vector |
Exponential and reciprocal
AVX-512 exponential and reciprocal instructions contain more accurate approximate reciprocal instructions than those in the AVX-512 foundation; relative error is at most 2−28. They also contain two new exponential functions that have a relative error of at most 2−23.| Instruction | Description |
VEXP2PD, VEXP2PS | Compute approximate exponential 2x of packed single or double-precision floating-point values |
VRCP28PD, VRCP28PS | Compute approximate reciprocals of packed single or double-precision floating-point values |
VRCP28SD, VRCP28SS | Compute approximate reciprocal of scalar single or double-precision floating-point value |
VRSQRT28PD, VRSQRT28PS | Compute approximate reciprocals of square roots of packed single or double-precision floating-point values |
VRSQRT28SD, VRSQRT28SS | Compute approximate reciprocal of square root of scalar single or double-precision floating-point value |
Prefetch
AVX-512 prefetch instructions contain new prefetch operations for the new scatter and gather functionality introduced in AVX2 and AVX-512.T0 prefetch means prefetching into level 1 cache and T1 means prefetching into level 2 cache.| Instruction | Description |
VGATHERPF0DPS, VGATHERPF0QPS,VGATHERPF0DPD, VGATHERPF0QPD | Using signed dword/qword indices, prefetch sparse byte memory locations containing single/double-precision data using opmask k1 and T0 hint. |
VGATHERPF1DPS, VGATHERPF1QPS,VGATHERPF1DPD, VGATHERPF1QPD | Using signed dword/qword indices, prefetch sparse byte memory locations containing single/double-precision data using opmask k1 and T1 hint. |
VSCATTERPF0DPS, VSCATTERPF0QPS,VSCATTERPF0DPD, VSCATTERPF0QPD | Using signed dword/qword indices, prefetch sparse byte memory locations containing single/double-precision data using writemask k1 and T0 hint with intent to write. |
VSCATTERPF1DPS, VSCATTERPF1QPS,VSCATTERPF1DPD, VSCATTERPF1QPD | Using signed dword/qword indices, prefetch sparse byte memory locations containing single/double precision data using writemask k1 and T1 hint with intent to write. |
4FMAPS and 4VNNIW
The two sets of instructions perform multiple iterations of processing. They are generally only found in Xeon Phi products.| Instruction | Extension set | Description |
V4FMADDPS,V4FMADDSS | 4FMAPS | Packed/scalar single-precision floating-point fused multiply-add |
V4FNMADDPS,V4FNMADDSS | 4FMAPS | Packed/scalar single-precision floating-point fused multiply-add and negate |
VP4DPWSSD | 4VNNIW | Dot product of signed words with double word accumulation |
VP4DPWSSDS | 4VNNIW | Dot product of signed words with double word accumulation and saturation |
BW, DQ and VBMI
AVX-512DQ adds new doubleword and quadword instructions. AVX-512BW adds byte and words versions of the same instructions, and adds byte and word version of doubleword/quadword instructions in AVX-512F. A few instructions which get only word forms with AVX-512BW acquire byte forms with the AVX-512_VBMI extension.Two new instructions were added to the mask instructions set:
KADD and KTEST. The rest of mask instructions, which had only word forms, got byte forms with AVX-512DQ and doubleword/quadword forms with AVX-512BW. KUNPCKBW was extended to KUNPCKWD and KUNPCKDQ by AVX-512BW.Among the instructions added by AVX-512DQ are several SSE and AVX instructions that didn't get AVX-512 versions with AVX-512F, among those are all the two input bitwise instructions and extract/insert integer instructions.
Instructions that are completely new are covered below.
Floating-point instructions
Three new floating-point operations are introduced. Since they are not only new to AVX-512 they have both packed/SIMD and scalar versions.The
VFPCLASS instructions tests if the floating-point value is one of eight special floating-point values, which of the eight values will trigger a bit in the output mask register is controlled by the immediate field. The VRANGE instructions perform minimum or maximum operations depending on the value of the immediate field, which can also control if the operation is done absolute or not and separately how the sign is handled. The VREDUCE instructions operate on a single source, and subtract from that the integer part of the source value plus a number of bits specified in the immediate field of its fraction.| Instruction | Extension set | Description |
VFPCLASSPS, VFPCLASSPD | DQ | Test types of packed single and double precision floating-point values. |
VFPCLASSSS, VFPCLASSSD | DQ | Test types of scalar single and double precision floating-point values. |
VRANGEPS, VRANGEPD | DQ | Range restriction calculation for packed floating-point values. |
VRANGESS, VRANGESD | DQ | Range restriction calculation for scalar floating-point values. |
VREDUCEPS, VREDUCEPD | DQ | Perform reduction transformation on packed floating-point values. |
VREDUCESS, VREDUCESD | DQ | Perform reduction transformation on scalar floating-point values. |
VBMI2
Extend VPCOMPRESS and VPEXPAND with byte and word variants. Shift instructions are new.| Instruction | Description |
VPCOMPRESSB, VPCOMPRESSW | Store sparse packed byte/word integer values into dense memory/register |
VPEXPANDB, VPEXPANDW | Load sparse packed byte/word integer values from dense memory/register |
VPSHLD | Concatenate and shift packed data left logical |
VPSHLDV | Concatenate and variable shift packed data left logical |
VPSHRD | Concatenate and shift packed data right logical |
VPSHRDV | Concatenate and variable shift packed data right logical |
VNNI
Vector Neural Network Instructions: AVX512-VNNI adds EVEX-coded instructions described below. With AVX-512F, these instructions can operate on 512-bit vectors, and AVX-512VL further adds support for 128- and 256-bit vectors.A later AVX-VNNI extension adds VEX encodings of these instructions which can only operate on 128- or 256-bit vectors. AVX-VNNI is not part of the AVX-512 suite, it does not require AVX-512F and can be implemented independently.
| Instruction | Description |
VPDPBUSD, VPDPBUSDS | Multiply groups of 4 pairs of signed bytes in the second input with corresponding unsigned bytes of the first input, summing those products and adding them to the doubleword result in the output. VPDPBUSDS performs signed saturation of the result. |
VPDPWSSD, VPDPWSSDS | Multiply groups of 2 pairs of signed words in the second input with corresponding signed words of the first input, summing those products and adding them to the doubleword result in the output. VPDPWSSDS performs signed saturation of the result. |
IFMA
Integer fused multiply-add instructions. AVX512-IFMA adds EVEX-coded instructions described below.A separate AVX-IFMA instruction set extension defines VEX encoding of these instructions. This extension is not part of the AVX-512 suite and can be implemented independently.
| Instruction | Extension set | Description |
VPMADD52LUQ | IFMA | Packed multiply of unsigned 52-bit integers and add the low 52-bit products to 64-bit accumulators |
VPMADD52HUQ | IFMA | Packed multiply of unsigned 52-bit integers and add the high 52-bit products to 64-bit accumulators |
GFNI
GFNI is a standalone instruction set extension and can be enabled separately from AVX or AVX-512. Depending on whether AVX and AVX-512F support is indicated by the CPU, GFNI support enables legacy, VEX or EVEX-coded instructions operating on 128, 256 or 512-bit vectors.In cryptography, these instructions can be used to implement Rijndael-style S-boxes, such as those used in Camellia and ARIA. These instructions can also be used for bit manipulation in networking and signal processing.
| Instruction | Description |
VGF2P8AFFINEINVQB | Galois field affine transformation inverse |
VGF2P8AFFINEQB | Galois field affine transformation |
VGF2P8MULB | Galois field multiply bytes |
VPCLMULQDQ
VPCLMULQDQ with AVX-512F adds an EVEX-encoded 512-bit version of the PCLMULQDQ instruction. With AVX-512VL, it adds EVEX-encoded 256- and 128-bit versions. VPCLMULQDQ alone adds only VEX-encoded 256-bit version. The wider than 128-bit variations of the instruction perform the same operation on each 128-bit portion of input registers, but they do not extend it to select quadwords from different 128-bit fields.| Instruction | Description |
VPCLMULQDQ | Carry-less multiplication quadword |
VAES
VEX- and EVEX-encoded AES instructions. The wider than 128-bit variations of the instruction perform the same operation on each 128-bit portion of input registers. The VEX versions can be used without AVX-512 support.| Instruction | Description |
VAESDEC | Perform one round of an AES decryption flow |
VAESDECLAST | Perform last round of an AES decryption flow |
VAESENC | Perform one round of an AES encryption flow |
VAESENCLAST | Perform last round of an AES encryption flow |
BF16
AI acceleration instructions operating on the Bfloat16 numbers.| Instruction | Description |
VCVTNE2PS2BF16 | Convert two vectors of packed single precision numbers into one vector of packed Bfloat16 numbers |
VCVTNEPS2BF16 | Convert one vector of packed single precision numbers to one vector of packed Bfloat16 numbers |
VDPBF16PS | Calculate dot product of two Bfloat16 pairs and accumulate the result into one packed single precision number |
FP16
An extension of the earlier F16C instruction set, adding comprehensive support for the binary16 floating-point numbers. The new instructions implement most operations that were previously available for single and double-precision floating-point numbers and also introduce new complex number instructions and conversion instructions. Scalar and packed operations are supported.Unlike the single and double-precision format instructions, the half-precision operands are neither conditionally flushed to zero nor conditionally treated as zero based on
MXCSR settings. Subnormal values are processed at full speed by hardware to facilitate using the full dynamic range of the FP16 numbers. Instructions that create FP32 and FP64 numbers still respect the MXCSR.FTZ bit.BMM
Bit matrix multiply and bit reversal instructions. Unlike other AVX-512 subsets, originally introduced by AMD.Bit matrix multiply instructions are only defined for 256 and 512-bit vectors, where each of the 256-bit lanes represent a 16x16 bit matrix. For 512-bit versions, each vector contains two matrices, and the operation is performed on them in parallel.
Bit reversal instruction is defined for 128, 256 and 512-bit vectors.
| Instruction | Description |
VBMACOR16x16x16 | 16x16 non-transposed fused BMM-accumulate with OR reduction. |
VBMACXOR16x16x16 | 16x16 non-transposed fused BMM-accumulate with XOR reduction. |
VBITREV | Bit reversal within a byte boundary. |
CPUs with AVX-512
- Intel
- * Knights Landing : AVX-512 F, CD, ER, PF
- * Knights Mill : AVX-512 F, CD, ER, PF, 4FMAPS, 4VNNIW, VPOPCNTDQ
- * Skylake-SP, Skylake-X: AVX-512 F, CD, VL, DQ, BW
- * Cannon Lake: AVX-512 F, CD, VL, DQ, BW, IFMA, VBMI
- * Cascade Lake: AVX-512 F, CD, VL, DQ, BW, VNNI
- * Cooper Lake: AVX-512 F, CD, VL, DQ, BW, VNNI, BF16
- * Ice Lake, Rocket Lake: AVX-512 F, CD, VL, DQ, BW, IFMA, VBMI, VBMI2, VPOPCNTDQ, BITALG, VNNI, VPCLMULQDQ, GFNI, VAES
- * Tiger Lake : AVX-512 F, CD, VL, DQ, BW, IFMA, VBMI, VBMI2, VPOPCNTDQ, BITALG, VNNI, VPCLMULQDQ, GFNI, VAES, VP2INTERSECT
- * Alder Lake : AVX-512 F, CD, VL, DQ, BW, IFMA, VBMI, VBMI2, VPOPCNTDQ, BITALG, VNNI, VPCLMULQDQ, GFNI, VAES, BF16, VP2INTERSECT, FP16
- * Sapphire Rapids and later P-core-only Xeon processors: AVX-512 F, CD, VL, DQ, BW, IFMA, VBMI, VBMI2, VPOPCNTDQ, BITALG, VNNI, VPCLMULQDQ, GFNI, VAES, BF16, FP16
- Centaur Technology
- * "CNS" core : AVX-512 F, CD, VL, DQ, BW, IFMA, VBMI
- AMD
- * Zen 4: AVX-512 F, CD, VL, DQ, BW, IFMA, VBMI, VBMI2, VPOPCNTDQ, BITALG, VNNI, VPCLMULQDQ, GFNI, VAES, BF16
- * Zen 5: AVX-512 F, CD, VL, DQ, BW, IFMA, VBMI, VBMI2, VPOPCNTDQ, BITALG, VNNI, VPCLMULQDQ, GFNI, VAES, BF16, VP2INTERSECT
- * Zen 6: AVX-512 F, CD, VL, DQ, BW, IFMA, VBMI, VBMI2, VPOPCNTDQ, BITALG, VNNI, VPCLMULQDQ, GFNI, VAES, BF16, VP2INTERSECT, FP16, BMM
In older Alder Lake family CPUs with some legacy combinations of BIOS and microcode revisions, it was possible to execute AVX-512 family instructions when disabling all the efficiency cores which do not contain the silicon for AVX-512.
Performance
Intel Vectorization Advisor supports native AVX-512 performance and vector code quality analysis. Along with traditional hotspots profile, Advisor Recommendations and "seamless" integration of Intel Compiler vectorization diagnostics, Advisor Survey analysis also provides AVX-512 ISA metrics and new AVX-512-specific "traits", e.g. Scatter, Compress/Expand, mask utilization.On some processors, AVX-512 instructions can cause a frequency throttling even greater than its predecessors, causing a penalty for mixed workloads. The additional downclocking is triggered by the 512-bit width of vectors and depends on the nature of instructions being executed; using the 128 or 256-bit part of AVX-512 does not trigger it. As a result, gcc and clang default to prefer using the 256-bit vectors for Intel targets.
C/C++ compilers also automatically handle loop unrolling and preventing stalls in the pipeline in order to use AVX-512 most effectively, which means a programmer using language intrinsics to try to force use of AVX-512 can sometimes result in worse performance relative to the code generated by the compiler when it encounters loops plainly written in the source code. In other cases, using AVX-512 intrinsics in C/C++ code can result in a performance improvement relative to plainly written C/C++.
Reception
There are many examples of AVX-512 applications, including media processing, cryptography, video games, neural networks, and even OpenJDK, which employs AVX-512 for sorting.In a much-cited quote from 2020, Linus Torvalds said "I hope AVX-512 dies a painful death, and that Intel starts fixing real problems instead of trying to create magic instructions to then create benchmarks that they can look good on", stating that he would prefer the transistor budget be spent on additional cores and integer performance instead, and that he "detests" floating point benchmarks.