Check out my first novel, midnight's simulacra!

SIMD: Difference between revisions

From dankwiki
(SIMDnxm notation)
 
(38 intermediate revisions by the same user not shown)
Line 1: Line 1:
==x86==
[[File:Simd.jpeg|thumb|right|A 4-way SIMD unit]]
* Terminology:
First introduced on the Illiac IV, and not to be confused with vector processing. SIMD-capable units are often taxonomized via <tt>SIMDnXm</tt> nomenclature, where '''n''' is the issue width and '''m''' is the channel width; a superscalar processor would be SIMDnX1, while a scalar SISD processor would be SIMD1X1.
** '''single''': 32-bit IEEE 754 floating-point (bias-127)
** '''double''': 64-bit IEEE 754 floating-point (bias-1023)
** '''long double''': 80-bit "double extended" IEEE 754-1985 floating-point (excess-16383)
*** not an actual SIMD type, but an artifact of x87
** '''word''': 32-bit two's complement integer
** '''doubleword''', '''dword''': 64-bit two's complement integer
* These do not necessarily map to the C data types of the same name, for any given compiler!


===Future===
==Compilers and Assemblers==
* [http://software.intel.com/en-us/avx/ AVX] (Advanced Vector eXtensions) -- to be introduced on Intel's Sandy Bridge (2010) and AMD's Bulldozer (2011), and implemented within the [http://en.wikipedia.org/wiki/VEX_prefix VEX coding scheme]
* [[GCC]] can issue SIMD four different ways:
* The [http://en.wikipedia.org/wiki/FMA_instruction_set FMA instruction set] extension to x86 should hit around 2011, providing floating-point fused multiply-add
** [http://gcc.gnu.org/onlinedocs/gcc/Submodel-Options.html#Submodel-Options Target-specific code generation] will sometimes use scalar SIMD
** [http://gcc.gnu.org/onlinedocs/gcc/Target-Builtins.html Target-specific builtins], using [http://gcc.gnu.org/onlinedocs/gcc/Vector-Extensions.html vector extensions]
** [[GCC#Inline_Assembly|Inline assembly]], using syntax of the [http://sourceware.org/binutils/docs/as/index.html GNU Assembler] and a baroque register specification convention
** [http://gcc.gnu.org/projects/tree-ssa/vectorization.html Auto-vectorization] from [[GCC|gcc's]] <tt>autovect</tt> branch attempts to autovectorize some loops
 
==Data Types (taken from SSE specs)==
These do not necessarily map to [[C]] data types of the same name, for any given compiler!
* '''half precision''': 16-bit IEEE 754 floating-point (bias-15) (IEEE 754 2008 '''binary16''')
* '''single''': 32-bit IEEE 754 floating-point (bias-127) (IEEE 754 2008 '''binary32''')
* '''double''': 64-bit IEEE 754 floating-point (bias-1023) (IEEE 754 2008 '''binary64''')
* '''long double''': 80-bit "double extended" IEEE 754-1985 floating-point (bias-16383)
** not an actual SIMD type, but an artifact of x87
* '''word''': 32-bit two's complement integer
* '''doubleword''', '''dword''': 64-bit two's complement integer
 
==x86 YMM==
The [http://software.intel.com/en-us/avx/ AVX] (Advanced Vector eXtensions) were introduced on Intel's [[Sandy Bridge]] (2010) and AMD's [[Bulldozer]] (2011). They operate on the 256-bit YMM registers (YMM0..YMM15), which are aliased by the [[SIMD#x86_XMM|XMM]] registers. They are encoded using the [[VEX]] scheme. Support for AVX can be determined via:
* Determine that [[CPUID]].1:ECX.OSXSAVE[bit 27] is set (XGETBV is enabled for application use)
* Issue XGETBV and verify that XFEATURE_ENABLED_MASK[2:1] = 0x3 (XMM state and YMM state are enabled by OS)
* Determine that [[CPUID]].1:ECX.AVX[bit 28] is set (AVX instructions supported)
==x86 AVX==
New 256-bit YMM registers.
===New [[Haswell]] Instructions===
* AVX2 instructions: integer SIMD using AVX's 256-bit YMM registers.
* YMM-SIMD bit manipulation, permute, gather, shift, and FMA
===AVX1===
* Introduced on [[Sandy Bridge]]
==x86 XMM==
The Streaming SIMD Extensions operate on the 128-bit XMM registers (XMM0..XMM7 in 32-bit mode, XMM0..XMM15 in 64-bit mode). In its original incarnation on the PIII, execution units (but not registers) were shared with the x87 floating-point architecture. The execution units were separated in the NetBurst microarchitecture. In the Core microarchitecture, the execution engine has been widened for greater SSE throughput.
===SSE5 (AMD)===
* Unimplemented extensions competing with SSE4, encoded using a method incompatible with VEX
* Withdrawn, converted into VEX-compatible encodings, and split into:
** FMA4: Fused floating-point multiply-add (compare Intel's FMA)
** XOP: Fused integer multiply-add, byte permutations, shifts, rotates, integer vector horizontal operations (compare Intel's SSE4)
** CVT16: Half-precision conversion


===SSE4===
===SSE4 (Intel)===
====SSE4a====
====SSE4.1====
====SSE4.1====
[[File:Dppd.gif|thumb|DPPD instruction dataflow]]
* Introduced on Penryn
*<tt>dpps</tt> -- dot product of two vectors having four single components each
*<tt>dppd</tt> -- dot product of two vectors having two double components each
*<tt>insertps</tt>
====SSE4.2====
====SSE4.2====
[[File:Intel Nehalem arch.png|thumb|Intel [[Nehalem]] microarchitecture]]
*Introduced on [[Nehalem]]
*<tt>crc32</tt>
*<tt>pcmpestri</tt>
*<tt>pcmpestrm</tt>
*<tt>pcmpistri</tt>
*<tt>pcmpistrm</tt>
*<tt>pcmpgtq</tt>
*<tt>popcnt</tt>


===SSE3===
===SSE4a (AMD)===
*<tt>lzcnt</tt>
*<tt>popcnt</tt>
*<tt>extrq</tt>
*<tt>insertq</tt>
*<tt>movntsd</tt>
*<tt>movntss</tt>
 
===SSE3 (PNI)===
*Originally known as Prescott New Instructions, and introduced on P4-Prescott
*<tt>[http://www.intel.com/software/products/documentation/vlin/mergedprojects/analyzer_ec/mergedprojects/reference_olh/mergedProjects/instructions/instruct32_hh/movddup--move_one_double-fp_and_duplicate.htm movddup]</tt> -- move a double from a 8-byte-aligned memory location or lower half of XMM register to upper half, then duplicate upper half to lower half
*<tt>[http://www.intel.com/software/products/documentation/vlin/mergedprojects/analyzer_ec/mergedprojects/reference_olh/mergedProjects/instructions/instruct32_hh/movddup--move_one_double-fp_and_duplicate.htm movddup]</tt> -- move a double from a 8-byte-aligned memory location or lower half of XMM register to upper half, then duplicate upper half to lower half
 
*<tt>MWAIT/MONITOR</tt>
====SSSE3====
====SSSE3 (TNI/MNI)====
*Introduced with the Core microarchitecture. Sometimes referred to as Tejas New Instructions or Merom New Instructions
*<tt>[http://www.sesp.cse.clrc.ac.uk/html/SoftwareTools/vtune/users_guide/mergedProjects/analyzer_ec/mergedProjects/reference_olh/mergedProjects/instructions/instruct32_hh/vc235.htm pmaddwd]</tt> -- multiply packed words, then horizontally sum pairs, accumulating into doublewords
*<tt>[http://www.sesp.cse.clrc.ac.uk/html/SoftwareTools/vtune/users_guide/mergedProjects/analyzer_ec/mergedProjects/reference_olh/mergedProjects/instructions/instruct32_hh/vc235.htm pmaddwd]</tt> -- multiply packed words, then horizontally sum pairs, accumulating into doublewords


===SSE2===
===SSE2===
*Introduced with the P4.
*Extends the [[SIMD#MMX|MMX]] instructions to XMM registers.
*<tt>[http://www.sesp.cse.clrc.ac.uk/html/SoftwareTools/vtune/users_guide/mergedProjects/analyzer_ec/mergedProjects/reference_olh/mergedProjects/instructions/instruct32_hh/vc180.htm movapd]</tt> -- move two packed doubles from a 16-byte-aligned memory location to XMM registers, or vice versa, or between two XMM registers.
*<tt>[http://www.sesp.cse.clrc.ac.uk/html/SoftwareTools/vtune/users_guide/mergedProjects/analyzer_ec/mergedProjects/reference_olh/mergedProjects/instructions/instruct32_hh/vc180.htm movapd]</tt> -- move two packed doubles from a 16-byte-aligned memory location to XMM registers, or vice versa, or between two XMM registers.
**<tt>[http://www.sesp.cse.clrc.ac.uk/html/SoftwareTools/vtune/users_guide/mergedProjects/analyzer_ec/mergedProjects/reference_olh/mergedProjects/instructions/instruct32_hh/vc205.htm movupd]</tt> -- <tt>movapd</tt> safe for unaligned memory references, with far inferior performance.
**<tt>[http://www.sesp.cse.clrc.ac.uk/html/SoftwareTools/vtune/users_guide/mergedProjects/analyzer_ec/mergedProjects/reference_olh/mergedProjects/instructions/instruct32_hh/vc205.htm movupd]</tt> -- <tt>movapd</tt> safe for unaligned memory references, with far inferior performance.
Line 31: Line 83:
*<tt>[http://www.intel.com/software/products/documentation/vlin/mergedprojects/analyzer_ec/mergedprojects/reference_olh/mergedprojects/instructions/instruct32_hh/vc8a.htm addpd]</tt> -- add two packed doubles. the addend is a 16-byte-aligned memory location or XMM register. the target XMM register serves as the augend.
*<tt>[http://www.intel.com/software/products/documentation/vlin/mergedprojects/analyzer_ec/mergedprojects/reference_olh/mergedprojects/instructions/instruct32_hh/vc8a.htm addpd]</tt> -- add two packed doubles. the addend is a 16-byte-aligned memory location or XMM register. the target XMM register serves as the augend.


===SSE===
===SSE (KNI/ISSE)===
* Introduced with the PIII. Sometimes referred to as Katmai New Instructions, and branded for some time as the Internet Streaming SIMD Extensions
*<tt>[http://www.sesp.cse.clrc.ac.uk/html/SoftwareTools/vtune/users_guide/mergedProjects/analyzer_ec/mergedProjects/reference_olh/mergedProjects/instructions/instruct32_hh/vc181.htm movaps]</tt> -- move four packed singles from a 16-byte-aligned memory location to XMM registers, or vice versa, or between two XMM registers.
*<tt>[http://www.sesp.cse.clrc.ac.uk/html/SoftwareTools/vtune/users_guide/mergedProjects/analyzer_ec/mergedProjects/reference_olh/mergedProjects/instructions/instruct32_hh/vc181.htm movaps]</tt> -- move four packed singles from a 16-byte-aligned memory location to XMM registers, or vice versa, or between two XMM registers.
**<tt>[http://www.sesp.cse.clrc.ac.uk/html/SoftwareTools/vtune/users_guide/mergedProjects/analyzer_ec/mergedProjects/reference_olh/mergedProjects/instructions/instruct32_hh/vc206.htm movups]</tt> -- <tt>movaps</tt> safe for unaligned memory references, with far inferior performance.
**<tt>[http://www.sesp.cse.clrc.ac.uk/html/SoftwareTools/vtune/users_guide/mergedProjects/analyzer_ec/mergedProjects/reference_olh/mergedProjects/instructions/instruct32_hh/vc206.htm movups]</tt> -- <tt>movaps</tt> safe for unaligned memory references, with far inferior performance.
*<tt>[http://www.sesp.cse.clrc.ac.uk/html/SoftwareTools/vtune/users_guide/mergedProjects/analyzer_ec/mergedProjects/reference_olh/mergedProjects/instructions/instruct32_hh/vc210.htm mulps]</tt> -- multiply four packed singles. the multiplier is a 16-byte-aligned memory location or XMM register. the target XMM register serves as the multiplicand.
*<tt>[http://www.sesp.cse.clrc.ac.uk/html/SoftwareTools/vtune/users_guide/mergedProjects/analyzer_ec/mergedProjects/reference_olh/mergedProjects/instructions/instruct32_hh/vc210.htm mulps]</tt> -- multiply four packed singles. the multiplier is a 16-byte-aligned memory location or XMM register. the target XMM register serves as the multiplicand.
*<tt>[http://www.sesp.cse.clrc.ac.uk/html/SoftwareTools/vtune/users_guide/mergedProjects/analyzer_ec/mergedProjects/reference_olh/mergedProjects/instructions/instruct32_hh/vc9a.htm addps]</tt> -- add four packed singles. the addend is a 16-byte-aligned memory location or XMM register. the target XMM register serves as the augend.
*<tt>[http://www.sesp.cse.clrc.ac.uk/html/SoftwareTools/vtune/users_guide/mergedProjects/analyzer_ec/mergedProjects/reference_olh/mergedProjects/instructions/instruct32_hh/vc9a.htm addps]</tt> -- add four packed singles. the addend is a 16-byte-aligned memory location or XMM register. the target XMM register serves as the augend.
===Future Directions===
* The [http://en.wikipedia.org/wiki/FMA_instruction_set FMA instruction set] extension to x86 should hit around 2011, providing floating-point fused multiply-add
** AMD appears to call this [http://en.wikipedia.org/wiki/FMA_instruction_set FMA4], part of what was SSE5
==x87 MMX==
===MMX (Intel)===
===3DNow! (AMD)===


==Other Architectures==
==Other Architectures==
* PowerPC implements [http://en.wikipedia.org/wiki/AltiVec AltiVec]
* PowerPC implements [http://en.wikipedia.org/wiki/AltiVec AltiVec]
* SPARC implements [http://en.wikipedia.org/wiki/Visual_Instruction_Set VIS], the Visual Instruction Set
* PA-RISC implements [http://en.wikipedia.org/wiki/Multimedia_Acceleration_eXtensions MAX], the Multimedia Acceleration eXtensions
* ARM implements [http://en.wikipedia.org/wiki/ARM_architecture#Advanced_SIMD_.28NEON.29 NEON]
* Alpha implemented [http://www.alphalinux.org/wiki/index.php/Motion_Video_Instructions MVI], the Motion Video Instructions
* [http://cobweb.ecn.purdue.edu/~hankd/SWAR/ SWAR]: SIMD Within a Register (bit-parallel methods)


==See Also==
==See Also==
* "[http://software.intel.com/en-us/forums/intel-avx-and-cpu-instructions/topic/61121/ Why no FMA in AVX in Sandy Bridge?]", Intel Developers Forum
* "[http://software.intel.com/en-us/forums/intel-avx-and-cpu-instructions/topic/61121/ Why no FMA in AVX in Sandy Bridge?]", Intel Developers Forum
* [http://developer.amd.com/cpu/SSE5/Pages/default.aspx SSE5] guide at AMD
* [http://developer.amd.com/cpu/SSE5/Pages/default.aspx SSE5] guide at AMD
* [http://softwarecommunity.intel.com/isn/Downloads/Intel%20SSE4%20Programming%20Reference.pdf SSE4] reference at Intel
* 2007-04-19 post to http://virtualdub.org, "[http://www.virtualdub.org/blog/pivot/entry.php?id=150 SSE4 finally adds dot products]"
* 2007-04-19 post to http://virtualdub.org, "[http://www.virtualdub.org/blog/pivot/entry.php?id=150 SSE4 finally adds dot products]"
 
* AMD64 Architecture Programmer’s Manual Volume 6: [http://support.amd.com/us/Processor_TechDocs/43479.pdf 128-Bit and 256-Bit XOP, FMA4 and CVT16 Instructions]
* Agner Fog's [http://www.agner.org/optimize/instruction_tables.pdf x86 Instruction Tables]
* "[http://software.intel.com/en-us/blogs/2011/06/13/haswell-new-instruction-descriptions-now-available Haswell New Instruction Descriptions now Available]", Intel Software Developer Network, 2011-06-13
* General [[architecture]] page
[[Category: x86]]
[[Category: x86]]

Latest revision as of 19:09, 17 February 2012

A 4-way SIMD unit

First introduced on the Illiac IV, and not to be confused with vector processing. SIMD-capable units are often taxonomized via SIMDnXm nomenclature, where n is the issue width and m is the channel width; a superscalar processor would be SIMDnX1, while a scalar SISD processor would be SIMD1X1.

Compilers and Assemblers

Data Types (taken from SSE specs)

These do not necessarily map to C data types of the same name, for any given compiler!

  • half precision: 16-bit IEEE 754 floating-point (bias-15) (IEEE 754 2008 binary16)
  • single: 32-bit IEEE 754 floating-point (bias-127) (IEEE 754 2008 binary32)
  • double: 64-bit IEEE 754 floating-point (bias-1023) (IEEE 754 2008 binary64)
  • long double: 80-bit "double extended" IEEE 754-1985 floating-point (bias-16383)
    • not an actual SIMD type, but an artifact of x87
  • word: 32-bit two's complement integer
  • doubleword, dword: 64-bit two's complement integer

x86 YMM

The AVX (Advanced Vector eXtensions) were introduced on Intel's Sandy Bridge (2010) and AMD's Bulldozer (2011). They operate on the 256-bit YMM registers (YMM0..YMM15), which are aliased by the XMM registers. They are encoded using the VEX scheme. Support for AVX can be determined via:

  • Determine that CPUID.1:ECX.OSXSAVE[bit 27] is set (XGETBV is enabled for application use)
  • Issue XGETBV and verify that XFEATURE_ENABLED_MASK[2:1] = 0x3 (XMM state and YMM state are enabled by OS)
  • Determine that CPUID.1:ECX.AVX[bit 28] is set (AVX instructions supported)

x86 AVX

New 256-bit YMM registers.

New Haswell Instructions

  • AVX2 instructions: integer SIMD using AVX's 256-bit YMM registers.
  • YMM-SIMD bit manipulation, permute, gather, shift, and FMA

AVX1

x86 XMM

The Streaming SIMD Extensions operate on the 128-bit XMM registers (XMM0..XMM7 in 32-bit mode, XMM0..XMM15 in 64-bit mode). In its original incarnation on the PIII, execution units (but not registers) were shared with the x87 floating-point architecture. The execution units were separated in the NetBurst microarchitecture. In the Core microarchitecture, the execution engine has been widened for greater SSE throughput.

SSE5 (AMD)

  • Unimplemented extensions competing with SSE4, encoded using a method incompatible with VEX
  • Withdrawn, converted into VEX-compatible encodings, and split into:
    • FMA4: Fused floating-point multiply-add (compare Intel's FMA)
    • XOP: Fused integer multiply-add, byte permutations, shifts, rotates, integer vector horizontal operations (compare Intel's SSE4)
    • CVT16: Half-precision conversion

SSE4 (Intel)

SSE4.1

DPPD instruction dataflow
  • Introduced on Penryn
  • dpps -- dot product of two vectors having four single components each
  • dppd -- dot product of two vectors having two double components each
  • insertps

SSE4.2

Intel Nehalem microarchitecture
  • Introduced on Nehalem
  • crc32
  • pcmpestri
  • pcmpestrm
  • pcmpistri
  • pcmpistrm
  • pcmpgtq
  • popcnt

SSE4a (AMD)

  • lzcnt
  • popcnt
  • extrq
  • insertq
  • movntsd
  • movntss

SSE3 (PNI)

  • Originally known as Prescott New Instructions, and introduced on P4-Prescott
  • movddup -- move a double from a 8-byte-aligned memory location or lower half of XMM register to upper half, then duplicate upper half to lower half
  • MWAIT/MONITOR

SSSE3 (TNI/MNI)

  • Introduced with the Core microarchitecture. Sometimes referred to as Tejas New Instructions or Merom New Instructions
  • pmaddwd -- multiply packed words, then horizontally sum pairs, accumulating into doublewords

SSE2

  • Introduced with the P4.
  • Extends the MMX instructions to XMM registers.
  • movapd -- move two packed doubles from a 16-byte-aligned memory location to XMM registers, or vice versa, or between two XMM registers.
    • movupd -- movapd safe for unaligned memory references, with far inferior performance.
  • mulpd -- multiply two packed doubles. the multiplier is a 16-byte-aligned memory location or XMM register. the target XMM register serves as the multiplicand.
  • addpd -- add two packed doubles. the addend is a 16-byte-aligned memory location or XMM register. the target XMM register serves as the augend.

SSE (KNI/ISSE)

  • Introduced with the PIII. Sometimes referred to as Katmai New Instructions, and branded for some time as the Internet Streaming SIMD Extensions
  • movaps -- move four packed singles from a 16-byte-aligned memory location to XMM registers, or vice versa, or between two XMM registers.
    • movups -- movaps safe for unaligned memory references, with far inferior performance.
  • mulps -- multiply four packed singles. the multiplier is a 16-byte-aligned memory location or XMM register. the target XMM register serves as the multiplicand.
  • addps -- add four packed singles. the addend is a 16-byte-aligned memory location or XMM register. the target XMM register serves as the augend.

Future Directions

  • The FMA instruction set extension to x86 should hit around 2011, providing floating-point fused multiply-add
    • AMD appears to call this FMA4, part of what was SSE5

x87 MMX

MMX (Intel)

3DNow! (AMD)

Other Architectures

  • PowerPC implements AltiVec
  • SPARC implements VIS, the Visual Instruction Set
  • PA-RISC implements MAX, the Multimedia Acceleration eXtensions
  • ARM implements NEON
  • Alpha implemented MVI, the Motion Video Instructions
  • SWAR: SIMD Within a Register (bit-parallel methods)

See Also