clarify basic principle

[libreriscv.git] / openpower / sv / rfc / ls001.mdwn

# OPF ISA WG External RFC LS001 v2 14Sep2022

* RFC Author: Luke Kenneth Casson Leighton.
* RFC Contributors/Ideas: Brad Frey, Paul Mackerras, Konstantinos Magritis,
  Cesar Strauss, Jacob Lifshay, Toshaan Bharvani, Dimitry Selyutin, Andrey
  Miroshnikov
* Funded by NLnet under the Privacy and Enhanced Trust Programme, EU
  Horizon2020 Grant 825310
* <https://git.openpower.foundation/isa/PowerISA/issues/64>
  [[ls001/discussion]]

This proposal is to extend the Power ISA with an Abstract RISC-Paradigm
Vectorisation Concept that may be orthogonally applied to **all and any** suitable
Scalar instructions, present and future, in the Scalar Power ISA.
The Vectorisation System is called
["Simple-V"](https://libre-soc.org/openpower/sv/)
and the Prefix Format is called
["SVP64"](https://libre-soc.org/openpower/sv/).
**Simple-V is not a Traditional Vector ISA and therefore
does not add Vector opcodes or regfiles**.
An ISA Concept similar to Simple-V was originally invented in 1994 by
Peter Hsu (Architect of the MIPS R8000) but was dropped as MIPS did not
have an Out-of-Order Microarchitecture at the time.

Simple-V is designed for Embedded Scenarios right the way through
Audio/Visual DSPs to 3D GPUs and Supercomputing.  As it does **not**
add actual Vector Instructions, relying solely and exclusively on the
**Scalar** ISA, it is **Scalar** instructions that need to be added to
the **Scalar** Power ISA before Simple-V may orthogonally Vectorise them.

The goal of RED Semiconductor Ltd, an OpenPOWER
Stakeholder, is to bring to market mass-volume general-purpose compute
processors that are competitive in the 3D GPU Audio Visual DSP EDGE IoT
desktop chromebook netbook smartphone laptop markets, performance-leveraged
by Simple-V.  Simple-V thus has to
be accompanied by corresponding **Scalar** instructions that bring the
**Scalar** Power ISA up-to-date.  These include IEEE754
[Transcendentals](https://libre-soc.org/openpower/transcendentals/)
[AV](https://libre-soc.org/openpower/sv/av_opcodes/)
cryptographic
[Biginteger](https://libre-soc.org/openpower/sv/biginteger/) and
[bitmanipulation](https://libre-soc.org/openpower/sv/bitmanip)
operations that ARM
Intel AMD and many other ISAs have been adding over the past 12 years
and Power ISA has not.  Three additional FP-related sets are needed
(missing from SFFS) -
[int_fp_mv](https://libre-soc.org/openpower/sv/int_fp_mv/)
[fclass](https://libre-soc.org/openpower/sv/fclass/) and
[fcvt](https://libre-soc.org/openpower/sv/fcvt/)
and one set named
[crweird](https://libre-soc.org/openpower/sv/cr_int_predication/)
increase the capability of CR Fields.

*Thus it becomes necesary to consider the Architectural Resource
Allocation of not just Simple-V but the 80-100 Scalar instructions all
at the same time*.

It is also critical to note that Simple-V **does not modify the Scalar
Power ISA**, that **only** Scalar words may be
Vectorised, and that Vectorised instructions are **not** permitted to be
different from their Scalar words (`addi` must use the same Word encoding
as `sv.addi`, and any new Prefixed instruction added **must** also
be added as Scalar).
The sole semi-exception is Vectorised
Branch Conditional, in order to provide the usual Advanced Branching
capability present in every Commercial 3D GPU ISA, but it
is the *Vectorised* Branch-Conditional that is augmented, not Scalar
Branch.

# Basic principle

The inspiration for Simple-V came from the fact that on examination of every
Vector ISA pseudocode encountered the Vector operations were expressed
as a for-loop on a Scalar element
operation, and then both a Scalar **and** a Vector instruction was added.
With Zero-Overhead Looping *already* being mainstream in DSPs for over three
decades it felt natural to separate the looping at both the ISA and
the Hardware Level
and thus provide only Scalar instructions (instantly halving the number
of instructions), but rather than go the VLIW route (TI MSP Series)
keep closely to existing Power ISA standard Scalar execution.

Thus the basic principle of Simple-V is to provide a Precise-Interruptible
Zero-Overhead Loop system[^zolc] with associated register "offsetting"
which augments a Suffixed instruction as a "template",
incrementing the register numbering progressively *and automatically*
each time round the "loop".  Thus it may be considered to be a form
of "Sub-Program-Counter" and at its simplest level can replace a large
sequence of regularly-increasing loop-unrolled instructions with just two:
one to set the Vector length and one saying where to
start from in the regfile.

On this sound and profoundly simple concept which leverages *Scalar*
Micro-architectural capabilities much more comprehensive festures are
easy to add, working up towards an ISA that easily matches the capability
of powerful 3D GPU Vector Supercomputing ISAs, without ever adding even
one single Vector opcode.

# Extension Levels

Simple-V has been subdivided into levels akin to the Power ISA Compliancy
Levels.  For now let us call them "SV Extension Levels" to differentiate
the two.  The reason for the
[SV Extension Levels](https://libre-soc.org/openpower/sv/compliancy_levels/)
is the same as for the
Power ISA Compliancy Levels (SFFS, SFS): to not overburden implementors
with features that they do not need.  *There is no dependence between
the two types of Levels*. The resources below therefore are
not all required for all SV Extension Levels but they are all required
to be reserved.

# Binary Interoperability

Power ISA has a reputation as being long-term stable.
**Simple-V guarantees binary interoperability** by defining fixed
register file bitwidths and size for all instructions.
The seduction of permitting different implementors to choose a register file
bitwidth and size with the same instructions unfortunately has
the catastrophic side-effect of introducing not only binary incompatibility
but silent data corruption as well as no means to trap-and-emulate differing
bitwidths.[^vsx256]

Thus "Silicon-Partner" Scalability
is prohibited in the Simple-V Scalable Vector ISA,
This does
mean that `RESERVED` space is crucial to have, in order
to safely provide the option of
future expanded register file bitwidths and sizes[^msr],
under explicitly-distinguishable encoding,
**at the discretion of and with the full authority of the OPF ISA WG**,
not the implementor ("Silicon Partner").

# Hardware Implementations

The fundamental principle of Simple-V is that it sits between Issue and
Decode, pausing the Program-Counter to service a "Sub-PC"
hardware for-loop.  This is very similar to "Zero-Overhead Loops"
in High-end DSPs (TI MSP Series).

Considerable effort has been expended to ensure that Simple-V is
practical to implement on an extremely wide range of Industry-wide
common **Scalar** micro-architectures.  Finite State Machine (for
ultra-low-resource and Mission-Critical), In-order single-issue, all the
way through to Great-Big Out-of-Order Superscalar Multi-Issue. The
SV Extension Levels specifically recognise these differing scenarios.

SIMD back-end ALUs particularly those with element-level predicate
masks may be exploited to good effect with very little additional
complexity to achieve high throughput, even on a single-issue in-order
microarchitecture. As usually becomes quickly apparent with in-order, its
limitations extend also to when Simple-V is deployed, which is why
Multi-Issue Out-of-Order is the recommended (but not mandatory) Scalar
Micro-architecture.

The only major concern is in the upper SV Extension Levels: the Hazard
Management for increased number of Scalar Registers to 128 (in current
versions) but given that IBM POWER9/10 has VSX register numbering 64,
and modern GPUs have 128, 256 amd even 512 registers this was deemed
acceptable. Strategies do exist in hardware for Hazard Management of
such large numbers of registers, even for Multi-Issue microarchitectures.

# Simple-V Architectural Resources

* No new Interrupt types are required.
  No modifications to existing Power ISA opcodes are required.
  No new Register Files are required (all because Simple-V is a category of
  Zero-Overhead Looping on Scalar instructions)
* GPR FPR and CR Field Register extend to 128.  A future
  version may extend to 256 or beyond[^extend] or also extend VSX[^futurevsx]
* 24-bits are needed within the main SVP64 Prefix (equivalent to a 2-bit XO)
* Another 24-bit (a second 2-bit XO) is needed for a planned future encoding,
  currently named "SVP64-Single"[^likeext001]
* A third 24-bits (third 2-bit XO) is strongly recommended to be `RESERVED`
  such that future unforeseen capability is needed (although this may be
  alternatively achieved with a mandatory PCR or MSR bit)
* To hold all Vector Context, five SPRs are needed for userspace.
  If Supervisor and Hypervisor mode are to
  also support Simple-V they will correspondingly need five SPRs each.
  (Some 32/32-to-64 aliases are advantageous but not critical).
* Five 6-bit XO (A-Form) "Management" instructions are needed.  These are
  Scalar 32-bit instructions and *may* be 64-bit-extended in future
  (safely within the SVP64 space: no need for an EXT001 encoding).

**Summary of Simple-V Opcode space**

* 75% of one Major Opcode (equivalent to the rest of EXT017)
* Five 6-bit XO 32-bit operations.

No further opcode space *for Simple-V* is envisaged to be required for
at least the next decade (including if added on VSX)

**Simple-V SPRs**

* **SVSTATE** - Vectorisation State sufficient for Precise-Interrupt
  Context-switching and no adverse latency.
* **SVSRR0** - identical in purpose to SRR0/1: storing SVSTATE on context-switch
  along-side MSR and PC.
* **SVSHAPE0-3** - these are 32-bit and may be grouped in pairs, they REMAP
  (shape) the Vectors
* **SVLR** - again similar to LR for exactly the same purpose, SVSTATE
  is swapped with SVLR by SV-Branch-Conditional for exactly the same
  reason that NIA is swapped with LR

**Vector Management Instructions**

These fit into QTY 5of 6-bit XO 32-bit encoding (svshape and svshape2 share
the same space):

* **setvl** - Cray-style Scalar Vector Length instruction
* **svstep** - used for Vertical-First Mode and for enquiring about internal
  state
* **svremap** - "tags" registers for activating REMAP
* **svshape** - convenience instruction for quickly setting up Matrix, DCT,
  FFT and Parallel Reduction REMAP
* **svshape2** - additional convenience instruction to set up "Offset" REMAP
  (fits within svshape's XO encoding)
* **svindex** - convenience instruction for setting up "Indexed" REMAP.

\newpage{}
# SVP64 24-bit Prefixes

The SVP64 24-bit Prefix (RM) provides several options,
all fitting within the 24-bit space (and no other).
These Modes do not interact with SVSTATE per se.  SVSTATE
primarily controls the looping (quantity, order), RM
influences the *elements* (the Suffix).  There is however
some close interaction when it comes to predication.
REMAP is outlined separately.
The primary options all of which are aimed at reducing instruction
count and reducing assembler complexity are:

* **element-width overrides**, which dynamically redefine each SFFS or SFS
  Scalar prefixed instruction to be 8-bit, 16-bit, 32-bit or 64-bit
  operands **without requiring new 8/16/32 instructions.**[^pseudorewrite]
  This results in full BF16 and FP16 opcodes being added to the Power ISA
  **without adding BF16 or FP16 opcodes** including full conversion
  between all formats.
* **predication**.
  this is an absolutely essential feature for a 3D GPU VPU ISA.
  CR Fields are available as Predicate Masks hence the reason for their
  extension to 128. Twin-Predication is also provided: this may best
  be envisaged as back-to-back VGATHER-VSCATTER but is not restricted
  to LD/ST, its use saves on instruction count.  Enabling one or other
  of the predicates provides all of the other types of operations
  found in Vector ISAs (VEXTRACT, VINSERT etc) again with no need
  to actually provide explicit such instructions.
* **Saturation**. **all** LD/ST and Arithmetic and Logical operations may
  be saturated (without adding explicit scalar saturated opcodes)
* **Reduction and Prefix-Sum** (Fibonnacci Series) Modes, including a
  "Reverse Gear".
* **vec2/3/4 "Packing" and "Unpacking"** (similar to VSX `vpack` and `vpkss`)
  accessible in a way that is easier than REMAP, added for the same reasons
  that drove `vpack` and `vpkss` etc. to be added: pixel, audio, and 3D
  data manipulation. With Pack/Unpack being part of SVSTATE it can be
  applied *in-place* saving register file space (no copy/mv needed).
* **Load/Store "fault-first"** speculative behaviour,
  identical to SVE and RVV
  Fault-first: provides auto-truncation of a speculative sequential parallel
  LD/ST batch, helping
  solve the "SIMD Considered Harmful" stripmining problem from a Memory
  Access perspective.
* **Data-Dependent Fail-First**: a 100% Deterministic extension of the LDST
  ffirst concept: first `Rc=1 BO test` failure terminates looping and 
  truncates VL to that exact point. Useful for implementing algorithms
  such as `strcpy` in around 14 high-performance Vector instructions, the
  option exists to include or exclude the failing element.
* **Predicate-result**: a strategic mode that effectively turns all and any
  operations into a type of `cmp`. An `Rc=1 BO test` is performed and if
  failing that element result is **not** written to the regfile. The `Rc=1`
  Vector of co-results **is** always written (subject to usual predication).
  Termed "predicate-result" because the combination of producing then
  testing a result is as if the test was in a follow-up predicated
  copy/mv operation, it reduces regfile pressure and instruction count.
  Also useful on saturated or other overflowing operations, the overflowing
  elements may be excluded from outputting to the regfile then
  post-analysed outside of critical hot-loops.

**RM Modes**

There are five primary categories of instructions in Power ISA, each of
which needed slightly different Modes. For example, saturation and
element-width overrides are meaningless to Condition Register Field
operations, and Reduction is meaningless to LD/ST but Saturation
saves register file ports in critical hot-loops. Thus the 24 bits may
be suitably adapted to each category.

* Normal - arithmetic and logical including IEEE754 FP
* LD/ST immediate - includes element-strided and unit-strided
* LD/ST indexed
* CR Field ops
* Branch-Conditional - saves on instruction count in 3D parallel if/else

It does have to be pointed out that there is huge pressure on the
Mode bits.  There was therefore insufficient room, unlike the way that
EXT001 was designed, to provide "identifying bits" *without first partially
decoding the Suffix*.  This should in no way be conflated with or taken
as an indicator that changing the meaning of the Suffix is performed
or desirable.

Some considerable care has been taken to ensure that Decoding may be
performed in a strict forward-pipelined fashion that, aside from changes in
SVSTATE (necessarily cached and propagated alongside MSR and PC)
and aside from the initial 32/64 length detection (also kept simple),
a Multi-Issue Engine would have no difficulty (performance maximisable).
With the initial partial RM Mode type-identification
decode performed above the Vector operations may then
easily be passed downstream in a fully forward-progressive piplined fashion
to independent parallel units for further analysis.

**Vectorised Branch-Conditional**

As mentioned in the introduction this is the one sole instruction group
that
is different pseudocode from its scalar equivalent. However even there
its various Mode bits and options can be set such that in the degenerate
case the behaviour becomes identical to Scalar Branch-Conditional.

The two additional Modes within Vectorised Branch-Conditional, both of
which may be combined, are `CTR-Mode` and `VLI-Test` (aka "Data Fail First").
CTR Mode extends the way that CTR may be decremented unconditionally
within Scalar Branch-Conditional, and not only makes it conditional but
also interacts with predication.  VLI-Test provides the same option
as Data-Dependent Fault-First to Deterministically truncate the Vector
Length at the fail **or success** point.

Boolean Logic rules on sets (treating the Vector of CR Fields to be tested by
`BO` as a set) dictate that the Branch should take place on either 'ALL'
tests succeeding (or failing) or whether 'SOME' tests succeed (or fail).
These options provide the ability to cover the majority of Parallel
3D GPU Conditions, saving a not inconsiderable number of instructions
especially given the close interaction with CTR in hot-loops. 

Also `SVLR` is introduced, which is a parallel twin of `LR`, and saving
and restoring of LR and SVLR may be deferred until the final decision
as to whether to branch.  In this way `sv.bclrl` does not corrupt `LR`.

Vectorised Branch-Conditional due to its side-effects (e.g. reducing CTR
or truncating VL) has practical uses even if the Branch is deliberately
set to the next instruction (CIA+8). For example it may be used to reduce
CTR by the number of bits set in a GPR, if that GPR is given as the predicate
mask `sv.bc/pm=r3`.

# SVP64Single 24-bits

The `SVP64-Single` 24-bit encoding focusses primarily on ensuring that
all 128 Scalar registers are fully accessible, provides element-width
overrides, one-bit predication
and brings Saturation to all existing Scalar operations.
BF16 and FP16 are thus
provided in the Scalar Power ISA without one single explicit FP16 or BF16
32-bit opcode being added.  The downside: such Scalar operations are
all 64-bit encodings.

As SVP64Single is new and still under development, space for it may
instead be `RESERVED`. It is however necessary in *some* form
as there are limitations
in SVP64 Register numbering, particularly for 4-operand instructions,
that can only be easily overcome by SVP64Single.

# Vertical-First Mode

This is a Computer Science term that needed first to be invented.
There exists only one other Vertical-First Vector ISA in the world:
Mitch Alsup's VVM Extension for the 66000, details of which may be
obtained publicly on `comp.arch` or directly from Mitch Alsup under
NDA. Several people have
independently derived Vertical-First: it simply did not have a
Computer Science term associated with it.

If we envisage register and Memory layout to be Horizontal and
instructions to be Vertical, and to then have some form of Loop
System (wherther Zero-Overhead or just branch-conditional based)
it is easier to then conceptualise VF vs HF Mode:

* Vertical-First progresses through *instructions* first before
  moving on to the next *register* (or Memory-address in the case
  of Mitch Alsup's VVM).
* Horizontal-First (also known as Cray-style Vectors) progresses
  through **registers** (or, register *elements* in traditional
  Cray-Vector ISAs) in full before moving on to the next *instruction*.

Mitch Alsup's VVM Extension is a form of hardware-level auto-vectorisation
based around Zero-Overhead Loops. Using a Variable-Length Encoding all
loop-invariant registers are "tagged" such that the Hazard Management
Engine may perform optimally and do less work in automatically identifying
parallelism opportunities.
With it not being appropriate to use Variable-Length Encoding in the Power
ISA a different much more explicit strategy was taken in Simple-V.

The biggest advantage inherent in Vertical-First is that it is very easy
to introduce into compilers, because all looping, as far as programs
is concerned, remains expressed as *Scalar assembler*.[^autovec]
Whilst Mitch Alsup's
VVM biggest strength is its hardware-level auto-vectorisation
but is limited in its ability to call
functions, Simple-V's Vertical-First provides explicit control over the
parallelism ("hphint")[^hphint] and also allows for full state to be stored/restored
(SVLR combined with LR), permitting full function calls to be made
from inside Vertical-First Loops, and potentially allows arbitrarily-depth
nested VF Loops.

Simple-V Vertical-First Looping requires an explicit instruction to
move `SVSTATE` regfile offsets forward: `svstep`. An early version of 
Vectorised
Branch-Conditional attempted to merge the functionality of `svstep`
into `sv.bc`: it became CISC-like in its complexity and was quickly reverted.

# Simple-V REMAP subsystem

[REMAP](https://libre-soc.org/openpower/sv/remap)
is extremely advanced but brings features already present in other
DSPs and Supercomputing ISAs. The usual sequential progression
through elements is pushed through a hardware-defined
*fully Deterministic*
"remapping".  Normally (without REMAP)
algorithms are costly or
convoluted to implement.  They are typically implemented
as hard-coded fully loop-unrolled assembler which is often
auto-generated by specialist tools, or written
entirely by hand.
All REMAP Schedules *including Indexed*
are 100% Deterministic from their point of declaration,
making it possible to forward-plan
Issue, Memory access and Register Hazard Management
in Multi-Issue Micro-architectures.
If combined with Vertical-First then much more complex operations may exploit
REMAP Schedules, such as Complex Number FFTs.

* **DCT/FFT** REMAP brings more capability than TI's MSP-Series DSPs and
  Qualcom Hexagon DSPs, and is not restricted to Integer or FP.
  (Galois Field is possible, implementing NTT). Operates *in-place*
  significantly reducing register usage.
* **Matrix** REMAP brings more capability than any other Matrix Extension
  (AMD GPUs, Intel, ARM), not being restricted to Power-2 sizes. Also not
  limited to the type of operation, it may perform Warshall Transitive
  Closure, Integer Matrix, Bitmanipulation Matrix, Galois Field (carryless
  mul) Matrix, and with care potentially Graph Maximum Flow as well. Also
  suited to Convolutions, Matrix Transpose and rotate, *all* of which is
  in-place.
* **General-purpose Indexed** REMAP, this option is provided to implement
  an equivalent of VSX `vperm`, as a general-purpose catch-all means of
  covering algorithms outside of the other REMAP Engines.
* **Parallel Reduction** REMAP, performs an automatic map-reduce using
  *any suitable scalar operation*.

Note that predication is possible on REMAP but is hard to use effectively.
It is often best to make copies of data (`VCOMPRESS`) then apply REMAP.

\newpage{}
# Scalar Operations

The primary reason for mentioning the additional Scalar operations
is because they are so numerous, with Power ISA not having advanced
in the *general purpose* compute area in the past 12 years, that some
considerable care is needed.

Summary:
**Including Simple-V, to fit everything at least 75% of 3 separate
Major Opcodes would be required**

Candidates (for all but the X-Form instructions) include:

* EXT006 (80% free)
* EXT017 (75% free but not recommended)
* EXT001 (50% free)
* EXT009 (100% free)
* EXT005 (100% free)
* brownfield space in EXT019 (25% but NOT recommended)

SVP64, SVP64-Single and SVP64-Reserved would require on their own each 25%
of one Major Opcode for a total of 75% of one Major Opcode. The remaining
**Scalar** opcodes, due to there being two separate sets of operations
with 16-bit immediates, will require the other space totalling two 75%
Majors.

Note critically that:

* Unlike EXT001, SVP64's 24-bits may **not** hold also any Scalar
  operations.  There is no free available space: a 25th bit would
  be required.  The entire 24-bits is **required** for the abstracted
  Hardware-Looping Concept **even when these 24-bits are zero**
* Any Scalar 64-bit instruction (regardless of how it is encoded) is unsafe to
  then Vectorise because this creates the situation of Prefixed-Prefixed,
  resulting in deep complexity in Hardware Decode at a critical juncture, as
  well as introducing 96-bit instructions.
* **All** of these Scalar instructions are candidates for Vectorisation.
  Thus none of them may be 64-bit-Scalar-only.

**Minor Opcodes to fit candidates above**

In order of size, for bitmanip and A/V DSP purposes:

* QTY 3of 2-bit XO: ternlogi, crternlogi, grevlogi
* QTY 7of 3-bit XO: xpermi, binlut, grevlog, swizzle-mv/fmv, bitmask, bmrevi
* QTY 8of 5/6-bit (A-Form): xpermi, bincrflut, bmask, fmvis, fishmv, bmrev,
  Galois Field
* QTY 30of 10-bit (X-Form): cldiv/mul, av-min/max/diff, absdac, xperm etc.
  (easily fit EXT019, EXT031).

Note: Some of the Galois Field operations will require QTY 1of Polynomial
SPR (per userspace supervisor hypervisor).

**EXT004**

For biginteger math, two instructions in the same space as "madd" are to
be proposed.  They are both 3-in 2-out operations taking or producing a
64-bit "pair" (like RTp), and perform 128/64 mul and div/mod operations
respectively.  These are **not** the same as VSX operations which are
128/128, and they are **not** the same as existing Scalar mul/div/mod,
all of which are 64/64 (or 64/32).

**EXT059 and EXT063**

Additionally for High-Performance Compute and Competitive 3D GPU, IEEE754 FP
Transcendentals are required, as are some DCT/FFT "Twin-Butterfly" operations.
For each of EXT059 and EXT063:

* QTY 33of X-Form "1-argument" (fsin, fsins, fcos, fcoss)
* QTY 15of X-Form "2-argument" (pow, atan2, fhypot)
* QTY 5of A-Form "3-in 2-out" FP Butterfly operations for DCT/FFT
* QTY 8of X-Form "2-in 2-out" FP Butterfly operations (again for DCT/FFT)
* An additional 16 instructions for IEEE754-2019
  (fminss/fmaxss, fminmag/fmaxmag)
  [under evaluation](https://bugs.libre-soc.org/show_bug.cgi?id=923)
  as of 08Sep2022

# Adding new opcodes.

With Simple-V being a type of Zero-Overhead Loop Engine on top of
Scalar operations some clear guidelines are needed on how both
existing "Defined Words" (Public v3.1 Section 1.6.3 term) and future
Scalar operations are added within the 64-bit space.  Examples of
legal and illegal allocations are given later.

The primary point is that once an instruction is defined in Scalar
32-bit form its corresponding space **must** be reserved in the
SVP64 area with the exact same 32-bit form, even if that instruction
is "Unvectoriseable" (`sc`, `sync`, `rfid` and `mtspr` for example).
Instructions may **not** be added in the Vector space without also
being added in the Scalar space, and vice-versa, *even if Unvectoriseable*.

This is extremely important because the worst possible situation
is if a conflicting Scalar instruction is added by another Stakeholder,
which then turns out to be Vectoriseable: it would then have to be
added to the Vector Space with a *completely different Defined Word*
and things go rapidly downhill in the Decode Phase from there.
Setting a simple inviolate rule helps avoid this scenario but does
need to be borne in mind when discussing potential allocation
schemes, as well as when new Vectoriseable Opcodes are proposed
for addition by future RFCs: the opcodes **must** be uniformly
added to Scalar **and** Vector spaces.

\newpage{}
# Potential Opcode allocation solution

There are unfortunately some inviolate requirements that directly place
pressure on the EXT000-EXT063 (32-bit) opcode space to such a degree that
it risks jeapordising the Power ISA. These requirements are:

* all of the scalar operations must be Vectoriseable
* all of the scalar operations intended for Vectorisation
  must be in a 32-bit encoding (not prefixed-prefixed to 96-bit)
* bringing Scalar Power ISA up-to-date from the past 12 years
  needs 75% of two Major opcodes all on its own

There exists a potential scheme which meets (exceeds) the above criteria,
providing plenty of room for both Scalar (and Vectorised) operations,
*and* provides SVP64-Single with room to grow.  It
is based loosely around Public v3.1 EXT001 Encoding.[^ext001]

| 0-5 | 6 | 7 | 8-31  | Description               |
|-----|---|---|-------|---------------------------|
| PO  | 0 | 0 | 0000  | new-suffix `RESERVED1`               |
| PO  | 0 | 0 | !zero | new-suffix, scalar (SVP64Single), or `RESERVED3` |
| PO  | 1 | 0 | 0000  | new scalar-only word, or `RESERVED2`               |
| PO  | 1 | 0 | !zero | old-suffix, scalar (SVP64Single), or `RESERVED4` |
| PO  | 0 | 1 | nnnn  | new-suffix, vector (SVP64)       |
| PO  | 1 | 1 | nnnn  | old-suffix, vector (SVP64)       |

* **PO** - Primary Opcode. Likely candidates: EXT005, EXT009
* **bit 6** - specifies whether the suffix is old (EXT000-EXT063)
  or new (EXTn00-EXTn63, n greater than 1)
* **bit 7** - defines whether the Suffix is Scalar-Prefixed or Vector-Prefixed
  (caveat: see bits 8-31)
* **old-suffix** - the EXT000 to EXT063 32-bit Major opcodes of Power ISA 3.0
* **new scalar-only** - a **new** Major Opcode area **exclusively**
  for Scalar-only instructions that shall **never** be Prefixed by SVP64
  (RESERVED2 EXT300-EXT363)
* **new-suffix** - a **new** Major Opcode area (RESERVED1 EXT200-EXT263)
  that **may** be Prefixed by SVP64 and SVP64Single
* **0000** - all 24 bits bits 8-31 are zero (0x000000)
* **!zero** - bits 8-31 may be any value *other* than zero (0x000001-0xffffff)
* **nnnn** - bits 8-31 may be any value in the range 0x000000 to 0xffffff
* **SVP64Single** - a ([TBD](https://bugs.libre-soc.org/show_bug.cgi?id=905))
  *Scalar* Encoding that is near-identical to SVP64
  except that it is equivalent to hard-coded VL=1
  at all times. Predication is permitted, Element-width-overrides is
  permitted, Saturation is permitted.
  If not allocated within the scope of this RFC
  then these are requested to be `RESERVED` for a future Simple-V
  proposal.
* **SVP64** - a (well-defined, 2 years) DRAFT Proposal for a Vectorisation
  Augmentation of suffixes.

For the needs identified by Libre-SOC (75% of 2 POs),
`RESERVED1` space *needs*
allocation to new POs, `RESERVED2` does not.[^only2]

|          | Scalar (bit7=0,8-31=0000) | Scalar (bit7=0,8-31=!zero)| Vector (bit7=1)  |
|----------|---------------------------|---------------------------|------------------|
|new bit6=0| `RESERVED1`:{EXT200-263}  | `RESERVED3`:SVP64-Single:{EXT200-263} | SVP64:{EXT200-263} |
|old bit6=1| `RESERVED2`:{EXT300-363}  | `RESERVED4`:SVP64-Single:{EXT000-063}   | SVP64:{EXT000-063}   |

* **`RESERVED2`:{EXT300-363}** (not strictly necessary to be added) is not
  and **cannot** ever be Vectorised or Augmented by Simple-V or any future
  Simple-V Scheme.
  it is a pure **Scalar-only** word-length PO Group. It may remain `RESERVED`.
* **`RESERVED1`:{EXT200-263}** is also a new set of 64 word-length Major
  Opcodes.
  These opcodes do not *need* to be Simple-V-Augmented
  *but the option to do so exists* should an Implementor choose to do so.
  This is unlike `EXT300-363` which may **never** be Simple-V-Augmented
  under any circumstances.
* **RESERVED3:`SVP64-Single:{EXT200-263}`** - Major opcodes 200-263 with
  Single-Augmentation, providing a one-bit predicate mask, element-width
  overrides on source and destination, and the option to extend the Scalar
  Register numbering (r0-32 extends to r0-127).  **Placing of alternative
  instruction encodings other than those exactly defined in EXT200-263
  is prohibited**.
* **RESERVED4:`SVP64-Single:{EXT000-063}`** - Major opcodes 000-063 with
  Single-Augmentation, just like SVP64-Single on EXT200-263, these are
  in effect Single-Augmented-Prefixed variants of the v3.0 32-bit Power ISA.
  Alternative instruction encodings other than the exact same 32-bit word
  from EXT000-EXT063 are likewise prohibited.
* **`SVP64:{EXT000-063}`** and **`SVP64:{EXT200-263}`** - Full Vectorisation
  of EXT000-063 and EXT200-263 respectively, these Prefixed instructions
  are likewise prohibited from being a different encoding from their
  32-bit scalar versions.

Limitations of this scheme is that new 32-bit Scalar operations have to have
a 32-bit "prefix pattern" in front of them. If commonly-used this could
increase binary size. Thus the Encodings EXT300-363 and EXT200-263 should
only be allocated for less-popular operations.  However the scheme does
have the strong advantage of *tripling* the available number of Major
Opcodes in the Power ISA, caveat being that care on allocation is needed
because EXT200-EXT263 may be SVP64-Augmented whilst EXT300-EXT363 may **not**.
The issues of allocation for bitmanip etc. from Libre-SOC is therefore
overwhelmingly made moot. The only downside is that there is no
`SVP64-Reserved` which will have to be achieved with SPRs (PCR or MSR).

# Potential Opcode allocation solution (2)

One of the risks of the bit 6/7 scheme above is that there is no
room to share PO9 (EXT009) with other potential uses.  A workaround for
that is as follows:

* EXT009, like EXT001 of Public v3.1, is **defined** as a 64-bit
  encoding
* bit 6 if 0b1 is 100% for Simple-V augmentation (Public v3.1 1.6.3)
  "Defined Word" (aka EXT000-063), with the exception of 0x24000000
  as a Prefix, which is a new RESERVED encoding.
* when bit 6 is 0b0 and bits 32-33 are 0b11 are **defined** as also
  allocated to Simple-V
* all other patterns are `RESERVED` for other purposes, 
  
| 0-5 | 6 | 7 | 8-31  | 32:33 |  Description               |
|-----|---|---|-------|-------|---------------------------|
| PO9?| 0 | 0 | 0000  | xx    | RESERVED (other)          |
| PO9?| 0 | 0 | !zero | 11    | SVP64 (current and future) |
| PO9?| 0 | 0 | !zero | 00-10 | RESERVED (other) |
| PO9?| 0 | 1 | xxxx  | 11    | SVP64 (current and future) |
| PO9?| 0 | x | xxxx  | 00-10 | RESERVED (other)          |
| PO9?| 1 | x | xxxx  | xx    | SVP64 (current and future) |

This ensures that any potential for future conflict over uses of the
EXT009 space, jeapordising Simple-V in the process, are avoided.

SVP64 then becomes:

| 0-5 | 6 | 7 | 8-31  | 32-3 | Description               |
|-----|---|---|-------|------|---------------------------|
| PO  | 0 | 0 | !zero | 0b11 | SVP64Single:EXT248-263, or `RESERVED3` |
| PO  | 0 | 0 | 0000  | 0b11 | Scalar EXT248-263               |
| PO  | 0 | 1 | nnnn  | 0b11 | SVP64:EXT248-263     |
| PO  | 1 | 0 | !zero | nn   | SVP64Single:EXT000-063 or `RESERVED4` |
| PO  | 1 | 1 | nnnn  | nn   | SVP64:EXT000-063       |

and reserved areas, QTY 1of 30-bit and QTY 3of 55-bit, are:

| 0-5 | 6 | 7 | 8-31  | 32-3 | Description               |
|-----|---|---|-------|------|---------------------------|
| PO  | 0 | 0 | 0000  | 0b11 | `RESERVED1` or EXT300-363      |
| PO9?| 0 | x | xxxx  | 0b00 | `RESERVED2` or EXT200-216    |
| PO9?| 0 | x | xxxx  | 0b01 | `RESERVED2` or EXT216-231    |
| PO9?| 0 | x | xxxx  | 0b10 | `RESERVED2` or EXT232-247   |

with additional potentially QTY 3of 30-bit reserved areas
(part of Scalar Unvectoriseable EXT200-247):

| 0-5 | 6 | 7 | 8-31  | 32-3 | Description               |
|-----|---|---|-------|------|---------------------------|
| PO9?| 1 | 0 | 0000  | 0b01 | RESERVED (other)          |
| PO9?| 1 | 0 | 0000  | 0b10 | RESERVED (other)          |
| PO9?| 1 | 0 | 0000  | 0b11 | RESERVED (other)          |

Where:

* SVP64Single (`RESERVED3/4`) is *planned* for a future RFC
  (but needs reserving as part of this RFC)
* `RESERVED1/2` is available for new general-purpose **never**-Simple-V
  (non-Vectoriseable) 32-bit encodings
* EXT248-263 is for "new" instructions
  which **must** also simultaneously request the corresponding space
  in SVP64, even if the instruction is non-Vectoriseable.
* Anything Vectorised-EXT000-063 is **automatically** being
  requested as 100% Reserved for every single "Defined Word"
  (Public v3.1 1.6.3 definition). Vectorised-EXT001 is defined as illegal.
* Any **future** instruction
  added to EXT000-063 likewise, is **automatically**
  assigned corresponding reservations in the SVP64:EXT000-063
  and SVP64Single:EXT000-063 area, regardless of whether the
  instruction is Vectoriseable or not.

Bit-allocation Summary:

* EXT3nn and three other encodings provide space for non-Simple-V
  operations to have QTY 4of EXTn00-EXTn47 Primary Opcode ranges
* QTY 3of 55-bit spaces also exist for future use (longer by 3 bits
  than opcodes allocated in EXT001)
* Simple-V EXT2nn is restricted to range EXT248-263
* non-Simple-V EXT2nn is restricted to range EXT200-247
* Simple-V EXT0nn takes up 50% of PO9 for this and future Simple-V RFCs
* The clear separation between Simple-V and non-Simple-V means there is
  no possibility of future RFCs encroaching on the others' space.

\newpage{}

**EXT000-EXT063**

These are Scalar word-encodings. Often termed "v3.0 Scalar" in this document
Power ISA v3.1 Section 1.6.3 Book I calls it a "defined word".

| 0-5    | 6-31   |
|--------|--------|
| PO     | EXT000-063 Scalar (v3.0 or v3.1) operation |

**SVP64Single:{EXT000-063}** bit6=old  bit7=scalar

This encoding, identical to SVP64Single:{EXT200-263},
introduces SVP64Single Augmentation of v3.0 Scalar word instructions.
All meanings must be identical to EXT000 to EXT063, and is is likewise
prohibited to add an instruction in this area without also adding
the exact same (non-Augmented) instruction in EXT000-063 with the
exact same Scalar word.
PO2 is in the range 0b00000 to 0b11111 to represent EXT000-063 respectively.
Augmenting EXT001 is prohibited.

| 0-5    | 6 | 7 | 8-31  | 32-63   |
|--------|---|---|-------|---------|
| PO (9)?| 1 | 0 | !zero | SVP64Single:{EXT000-063} |

**SVP64:{EXT000-063}** bit6=old bit7=vector

This encoding is identical to **SVP64:{EXT200-263}** except it
is the Vectorisation of existing v3.0/3.1 Scalar-words, EXT000-063.
All the same rules apply with the addition that
Vectorisation of EXT001 is prohibited.

| 0-5    | 6 | 7 | 8-31  | 32-63   |
|--------|---|---|-------|---------|
| PO (9)?| 1 | 1 | nnnn  | SVP64:{EXT000-063} |

**{EXT248-263}** bit6=new bit7=scalar

This encoding represents the opportunity to introduce EXT248-263.
It is a Scalar-word encoding, and does not require implementing
SVP64 or SVP64-Single, but does require the Vector-space to be allocated.
PO2 is in the range 0b11000 to 0b111111 to represent EXT248-263 respectively.

| 0-5    | 6 | 7 | 8-31  | 32-3 | 34-37   | 38-63   |
|--------|---|---|-------|------|---------|---------|
| PO (9)?| 0 | 0 | 0000  | 0b11 |PO2[2:5] | {EXT248-263} |

**SVP64Single:{EXT248-263}** bit6=new bit7=scalar

This encoding, which is effectively "implicit VL=1"
and comprising (from bits 8-31)
*at least some* form of Augmentation, it represents the opportunity
to Augment EXT248-263 with the SVP64Single capabilities.
Must be allocated under Scalar *and* SVP64 simultaneously.

| 0-5    | 6 | 7 | 8-31  | 32-3 | 34-37   | 38-63   |
|--------|---|---|-------|------|---------|---------|
| PO (9)?| 0 | 0 | !zero | 0b11 |PO2[2:5] | SVP64Single:{EXT248-263} |

**SVP64:{EXT248-263}** bit6=new bit7=vector

This encoding, which permits VL to be dynamic (settable from GPR or CTR)
is the Vectorisation of EXT248-263.
Instructions may not be placed in this category without also being
implemented as pure Scalar *and* SVP64Single. Unlike SVP64Single
however, there is **no reserved encoding** (bits 8-24 zero).
VL=1 may occur dynamically
at runtime, even when bits 8-31 are zero.

| 0-5    | 6 | 7 | 8-31  | 32-3 | 34-37   | 38-63   |
|--------|---|---|-------|------|---------|---------|
| PO (9)?| 0 | 1 | nnnn  | 0b11 |PO2[2:5] | SVP64:{EXT248-263} |

**RESERVED2 / EXT300-363** bit6=old bit7=scalar

This is entirely at the discretion of the ISA WG. Libre-SOC is *not*
proposing the addition of EXT300-363: it is merely a possibility for
future.  The reason the space is not needed is because this is within
the realm of Scalar-extended (SVP64Single), and with the 24-bit prefix
area being all-zero (bits 8-31) this is defined as "having no augmentation"
(in the Simple-V Specification it is termed `Scalar Identity Behaviour`).
This in turn makes this prefix a *degenerate duplicate* so may be allocated
for other purposes.

| 0-5    | 6 | 7 | 8-31  | 32-63   |
|--------|---|---|-------|---------|
| PO (9)?| 1 | 0 | 0000  | EXT300-363 or `RESERVED1` |

\newpage{}
# Example Legal Encodings and RESERVED spaces

This section illustrates what is legal encoding, what is not, and
why the 4 spaces should be `RESERVED` even if not allocated as part
of this RFC.

**legal, scalar and vector**

| width | assembler | prefix?      | suffix    | description   |
|-------|-----------|--------------|-----------|---------------|
| 32bit | fishmv    | none         | 0x12345678| scalar EXT0nn |
| 64bit | ss.fishmv | 0x26!zero    | 0x12345678| scalar SVP64Single:EXT0nn |
| 64bit | sv.fishmv | 0x27nnnnnn   | 0x12345678| vector SVP64:EXT0nn |

OR:

| width | assembler | prefix?      | suffix    | description   |
|-------|-----------|--------------|-----------|---------------|
| 64bit | fishmv    | 0x24000000   | 0x12345678| scalar EXT2nn |
| 64bit | ss.fishmv | 0x24!zero    | 0x12345678| scalar SVP64Single:EXT2nn |
| 64bit | sv.fishmv | 0x25nnnnnn   | 0x12345678| vector SVP64:EXT2nn |

Here the encodings are the same, 0x12345678 means the same thing in 
all cases. Anything other than this risks either damage (truncation
of capabilities of Simple-V) or far greater complexity in the
Decode Phase.

This drives the compromise proposal (above) to reserve certain
EXT2nn POs right
across the board
(in the Scalar Suffix side, irrespective of Prefix), some allocated
to Simple-V, some not.  

**illegal due to missing**

| width | assembler | prefix?      | suffix    | description   |
|-------|-----------|--------------|-----------|---------------|
| 32bit | fishmv    | none         | 0x12345678| scalar EXT0nn |
| 64bit | ss.fishmv | 0x26!zero    | 0x12345678| scalar SVP64Single:EXT0nn |
| 64bit | unallocated | 0x27nnnnnn   | 0x12345678| vector SVP64:EXT0nn |

This is illegal because the instruction is possible to Vectorise,
therefore it should be **defined** as Vectoriseable.

**illegal due to unvectoriseable**

| width | assembler | prefix?      | suffix    | description   |
|-------|-----------|--------------|-----------|---------------|
| 32bit | mtmsr     | none         | 0x12345678| scalar EXT0nn |
| 64bit | ss.mtmsr  | 0x26!zero    | 0x12345678| scalar SVP64Single:EXT0nn |
| 64bit | sv.mtmsr  | 0x27nnnnnn   | 0x12345678| vector SVP64:EXT0nn |

This is illegal because the instruction `mtmsr` is not possible to Vectorise,
at all.  This does **not** convey an opportunity to allocate the
space to an alternative instruction.

**illegal unvectoriseable in EXT2nn**

| width | assembler | prefix?      | suffix    | description   |
|-------|-----------|--------------|-----------|---------------|
| 64bit | mtmsr2    | 0x24000000   | 0x12345678| scalar EXT2nn |
| 64bit | ss.mtmsr2 | 0x24!zero    | 0x12345678| scalar SVP64Single:EXT2nn |
| 64bit | sv.mtmsr2 | 0x25nnnnnn   | 0x12345678| vector SVP64:EXT2nn |

For a given hypothetical `mtmsr2` which is inherently Unvectoriseable
whilst it may be put into the scalar EXT2nn space it may **not** be
allocated in the Vector space. As with Unvectoriseable EXT0nn opcodes
this does not convey the right to use the 0x24/0x26 space for alternative
opcodes.  This hypothetical Unvectoriseable operation would be better off
being allocated as EXT001 Prefixed, EXT000-063, or hypothetically in
EXT300-363.

**ILLEGAL: dual allocation**

| width | assembler | prefix?      | suffix    | description   |
|-------|-----------|--------------|-----------|---------------|
| 32bit | fredmv    | none         | 0x12345678| scalar EXT0nn |
| 64bit | ss.fredmv | 0x26!zero    | 0x12345678| scalar SVP64Single:EXT0nn |
| 64bit | sv.fishmv  | 0x27nnnnnn   | 0x12345678| vector SVP64:EXT0nn |

the use of 0x12345678 for fredmv in scalar but fishmv in Vector is
illegal.  the suffix in both 64-bit locations
must be allocated to a Vectoriseable EXT000-063
"Defined Word" (Public v3.1 Section 1.6.3 definition)
or not at all.

**illegal unallocated scalar EXT0nn or EXT2nn:**

| width | assembler | prefix?      | suffix    | description   |
|-------|-----------|--------------|-----------|---------------|
| 32bit | unallocated | none         | 0x12345678| scalar EXT0nn |
| 64bit | ss.fredmv | 0x26!zero    | 0x12345678| scalar SVP64Single:EXT0nn |
| 64bit | sv.fishmv  | 0x27nnnnnn   | 0x12345678| vector SVP64:EXT0nn |

and:

| width | assembler | prefix?      | suffix    | description   |
|-------|-----------|--------------|-----------|---------------|
| 64bit | unallocated | 0x24000000 | 0x12345678| scalar EXT2nn |
| 64bit | ss.fishmv | 0x24!zero    | 0x12345678| scalar SVP64Single:EXT2nn |
| 64bit | sv.fishmv | 0x25nnnnnn   | 0x12345678| vector SVP64:EXT2nn |

Both of these Simple-V operations are illegally-allocated. The fact that
there does not exist a scalar "Defined Word" (even for EXT200-263) - the
unallocated block - means that the instruction may **not** be allocated in
the Simple-V space.  

\newpage{}
# Use cases

In the following examples the programs are fully executable under the
Libre-SOC Simple-V-augmented Power ISA Simulator.  Reproducible
(scripted) Installation instructions:
<https://libre-soc.org/HDL_workflow/devscripts/>

## LD/ST-Multi

Context-switching saving and restoring of registers on the stack often
requires explicit loop-unrolling to achieve effectively.  In SVP64 it
is possible to use a Predicate Mask to "compact" or "expand" a swathe
of desired registers, dynamically.  Known as "VCOMPRESS" and "VEXPAND",
runtime-configurable LD/ST-Multi is achievable with 2 instructions.

```
    # load 64 registers off the stack, in-order, skipping unneeded ones
    # by using CR0-CR63's "EQ" bits to select only those needed.
    setvli 64
    sv.ld/sm=EQ *rt,0(ra)
```

## Twin-Predication, re-entrant

This example demonstrates two key concepts: firstly Twin-Predication
(separate source predicate mask from destination predicate mask) and
that sufficient state is stored within the Vector Context SPR, SVSTATE,
for full re-entrancy on a Context Switch or function call *even if
in the middle of executing a loop*.  Also demonstrates that it is
permissible for a programmer to write **directly** to the SVSTATE
SPR, and still expect Deterministic Behaviour. It's not exactly recommended
(performance may be impacted by direct SVSTATE access), but it is not
prohibited either.

```
 292     # checks that we are able to resume in the middle of a VL loop,
 293     # after an interrupt, or after the user has updated src/dst step
 294     # let's assume the user has prepared src/dst step before running this
 295     # vector instruction
 296     # test_intpred_reentrant
 297         #   reg num        0 1 2 3 4 5 6 7 8 9 10 11 12
 298         #   srcstep=1                           v
 299         #   src r3=0b0101                    Y  N  Y  N
 300         #                                    :     |
 301         #                              + - - +     |
 302         #                              :   +-------+
 303         #                              :   |
 304         #   dest ~r3=0b1010          N Y N Y
 305         #   dststep=2                    ^
 306
 307         sv.extsb/sm=r3/dm=~r3 *5, *9
```

<https://git.libre-soc.org/?p=openpower-isa.git;a=blob;f=src/openpower/decoder/isa/test_caller_svp64_predication.py;hb=HEAD>

## 3D GPU style "Branch Conditional"

(*Note: Specification is ready, Simulator still under development of
full specification capabilities*)
This example demonstrates a 2-long Vector Branch-Conditional only
succeeding if *all* elements in the Vector are successful.  This
avoids the need for additional instructions that would need to
perform a Parallel Reduction of a Vector of Condition Register
tests down to a single value, on which a Scalar Branch-Conditional
could then be performed.  Full Rationale at
<https://libre-soc.org/openpower/sv/branches/>

```
  80   # test_sv_branch_cond_all
  81       for i in [7, 8, 9]:
  83               addi 1, 0, i+1        # set r1 to i
  84               addi 2, 0, i          # set r2 to i
  85               cmpi cr0, 1, 1, 8     # compare r1 with 10 and store to cr0
  86               cmpi cr1, 1, 2, 8     # compare r2 with 10 and store to cr1
  87               sv.bc/all 12, *1, 0xc # bgt 0xc - branch if BOTH
  88                                     # r1 AND r2 greater 8 to the nop below
  89               addi 3, 0, 0x1234,    # if tests fail this shouldn't execute
  90               or 0, 0, 0            # branch target
```

<https://git.libre-soc.org/?p=openpower-isa.git;a=blob;f=src/openpower/decoder/isa/test_caller_svp64_bc.py;hb=HEAD>

\newpage{}
## DCT

DCT has dozens of uses in Audio-Visual processing and CODECs.
A full 8-wide in-place triple-loop Inverse DCT may be achieved
in 8 instructions.  Expanding this to 16-wide is a matter of setting
`svshape 16` **and the same instructions used**.
Lee Composition may be deployed to construct non-power-two DCTs.
The cosine table may be computed (once) with 18 Vector instructions
(one of them `fcos`)

```
1014     # test_sv_remap_fpmadds_ldbrev_idct_8_mode_4
1015                       # LOAD bit-reversed with half-swap
1016                       svshape 8, 1, 1, 14, 0
1017                       svremap 1, 0, 0, 0, 0, 0, 0
1018                       sv.lfs/els *0, 4(1)
1019                       # Outer butterfly, iterative sum
1020                       svremap 31, 0, 1, 2, 1, 0, 1
1021                       svshape 8, 1, 1, 11, 0
1022                       sv.fadds *0, *0, *0
1023                       # Inner butterfly, twin +/- MUL-ADD-SUB
1024                       svshape 8, 1, 1, 10, 0
1025                       sv.ffmadds *0, *0, *0, *8
```

<https://git.libre-soc.org/?p=openpower-isa.git;a=blob;f=src/openpower/decoder/isa/test_caller_svp64_dct.py;hb=HEAD>

## Matrix Multiply

Matrix Multiply of any size (non-power-2) up to a total of 127 operations
is achievable with only three instructions.  Normally in any other SIMD
ISA at least one source requires Transposition and often massive rolling
repetition of data is required.  These 3 instructions may be used as the
"inner triple-loop kernel" of the usual 6-loop Massive Matrix Multiply.

```
  28     # test_sv_remap1   5x4 by 4x3 matrix multiply
  29                        svshape 5, 4, 3, 0, 0
  30                        svremap 31, 1, 2, 3, 0, 0, 0
  31                        sv.fmadds *0, *8, *16, *0
```

<https://git.libre-soc.org/?p=openpower-isa.git;a=blob;f=src/openpower/decoder/isa/test_caller_svp64_matrix.py;hb=HEAD>

## Parallel Reduction

Parallel (Horizontal) Reduction is often deeply problematic in SIMD and
Vector ISAs.  Parallel Reduction is Fully Deterministic in Simple-V and
thus may even usefully be deployed on non-associative and non-commutative
operations.

```
  75     # test_sv_remap2
  76                        svshape 7, 0, 0, 7, 0
  77                        svremap 31, 1, 0, 0, 0, 0, 0 # different order
  78                        sv.subf *0, *8, *16
  79
  80                 REMAP sv.subf RT,RA,RB - inverted application of RA/RB
  81                                          left/right due to subf
```

<https://git.libre-soc.org/?p=openpower-isa.git;a=blob;f=src/openpower/decoder/isa/test_caller_svp64_parallel_reduce.py;hb=HEAD>

[[!tag opf_rfc]]

[^zolc]: first introduced in DSPs, Zero-Overhead Loops are astoundingly effective in reducing total number of instructions executed or needed. [ZOLC](https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.301.4646&rep=rep1&type=pdf) reduces instructions by **25 to 80 percent**.
[^msr]: an MSR bit or bits, conceptually equivalent to `MSR.SF` and added for the same reasons, would suffice perfectly.
[^extend]: Prefix opcode space (or MSR bits) **must** be reserved in advance to do so, in order to avoid the catastrophic binary-incompatibility mistake made by RISC-V RVV and ARM SVE/2
[^likeext001]: SVP64-Single is remarkably similar to the "bit 1" of EXT001 being set to indicate that the 64-bits is to be allocated in full to a new encoding, but in fact SVP64-single still embeds v3.0 Scalar operations.
[^pseudorewrite]: elwidth overrides does however mean that all SFS / SFFS pseudocode will need rewriting to be in terms of XLEN. This has the indirect side-effect of automatically making a 32-bit Scalar Power ISA Specification possible, as well as a future 128-bit one (Cross-reference: RISC-V RV32 and RV128
[^only2]: reminder that this proposal only needs 75% of two POs for Scalar instructions. The rest of EXT200-263 is for general use.
[^ext001]: Recall that EXT100 to EXT163 is for Public v3.1 64-bit-augmented Operations prefixed by EXT001, for which, from Section 1.6.3, bit 6 is set to 1.  This concept is where the above scheme originated. Section 1.6.3 uses the term "defined word" to refer to pre-existing EXT000-EXT063 32-bit instructions so prefixed to create the new numbering EXT100-EXT163, respectively
[^futurevsx]: A future version or other Stakeholder *may* wish to drop Simple-V onto VSX: this would be a separate RFC
[^vsx256]: imagine a hypothetical future VSX-256 using the exact same instructions as VSX. the binary incompatibility introducrd would catastrophically **and retroactively** damage existing IBM POWER8,9,10 hardware's reputation and that of Power ISA overall.
[^autovec]: Compiler auto-vectorisation for best exploitation of SIMD and Vector ISAs on Scalar programming languages (c, c++) is an Indusstry-wide known-hard decades-long problem. Cross-reference the number of hand-optimised assembler algorithms.
[^hphint]: intended for use when the compiler has determined the extent of Memory or register aliases in loops: `a[i] += a[i+4]` would necessitate a Vertical-First hphint of 4