1 # Condition Register SVP64 Operations
7 * <https://bugs.libre-soc.org/show_bug.cgi?id=687>
8 * <https://bugs.libre-soc.org/show_bug.cgi?id=936> write on failfirst
11 * [[sv/cr_int_predication]]
12 * [[openpower/isa/sprset]]
13 * [[openpower/isa/condition]]
14 * [[openpower/isa/comparefixed]]
16 Condition Register Fields are only 4 bits wide: this presents some
17 interesting conceptual challenges for SVP64, which was designed
18 primarily for vectors of arithmetic and logical operations. However
19 if predicates may be bits of CR Fields it makes sense to extend
20 Simple-V to cover CR Operations, especially given that Vectorized Rc=1
21 may be processed by Vectorized CR Operations that usefully in turn
22 may become Predicate Masks to yet more Vector operations, like so:
25 sv.cmpi/ew=8 *B,*ra,0 # compare bytes against zero
26 sv.cmpi/ew=8 *B2,*ra,13. # and against newline
27 sv.cror PM.EQ,B.EQ,B2.EQ # OR compares to create mask
28 sv.stb/sm=EQ ... # store only nonzero/newline
31 Element width however is clearly meaningless for a 4-bit collation of
32 Conditions, EQ LT GE SO. Likewise, arithmetic saturation (an important
33 part of Arithmetic SVP64) has no meaning. An alternative Mode Format is
34 required, and given that elwidths are meaningless for CR Fields the bits
35 in SVP64 `RM` may be used for other purposes.
37 This alternative mapping **only** applies to instructions that **only**
38 reference a CR Field or CR bit as the sole exclusive result. This section
39 **does not** apply to instructions which primarily produce arithmetic
40 results that also, as an aside, produce a corresponding CR Field (such as
41 when Rc=1). Instructions that involve Rc=1 are definitively arithmetic
42 in nature, where the corresponding Condition Register Field can be
43 considered to be a "co-result". Such CR Field "co-result" arithmeric
44 operations are firmly out of scope for this section, being covered fully
47 * Examples of Vectorizeable Defined Word-instructions to which this section does
49 - `mfcr` and `cmpi` (3 bit operands) and
50 - `crnor` and `crand` (5 bit operands).
51 * Examples to which this section does **not** apply include
52 `fadds.` and `subf.` which both produce arithmetic results
53 (and a CR Field co-result).
54 * `mtcr` is considered [[openpower/sv/normal]] because it refers
55 to the entire 32-bit Condition Register rather than to CR Fields.
57 The CR Mode Format still applies to `sv.cmpi` because despite
58 taking a GPR as input, the output from the Base Scalar v3.0B `cmpi`
59 instruction is purely to a Condition Register Field.
61 Other modes are still applicable and include:
63 * **Data-dependent fail-first**.
64 useful to truncate VL based on analysis of a Condition Register result bit.
66 Reduction is useful for analysing a Vector of Condition Register Fields
67 and reducing it to one single Condition Register Field.
69 Special atrention should be paid on the difference between Data-Dependent Fail-First
70 on CR operations and [[openpower/sv/normal]] regarding the seemingly-contradictory
71 behaviour of `Rc=1,VLi=0`. Explained below.
75 SVP64 RM `MODE` (includes `ELWIDTH_SRC` bits) for CR-based operations:
77 |6 | 7 |19:20|21 | 22:23 | description |
78 |--|---|-----|---|---------|------------------|
79 |/ | / |0 0 |RG | dz sz | simple mode |
80 |/ | / |1 0 |RG | dz sz | scalar reduce mode (mapreduce) |
81 |zz|SNZ|VLI 1|inv| CR-bit | Ffirst 3-bit mode |
82 |/ |SNZ|VLI 1|inv| dz sz | Ffirst 5-bit mode (implies CR-bit from result) |
86 * **sz / dz** if predication is enabled will put zeros into the dest
87 (or as src in the case of twin pred) when the predicate bit is zero.
88 otherwise the element is ignored or skipped, depending on context.
89 * **zz** set both sz and dz equal to this flag
90 * **SNZ** In fail-first mode, on the bit being tested, when sz=1 and
91 SNZ=1 a value "1" is put in place of "0".
92 * **inv CR-bit** just as in branches (BO) these bits allow testing of
93 a CR bit and whether it is set (inv=0) or unset (inv=1)
94 * **RG** Reverse-Gear: inverts the Vector Loop order (VL-1 downto 0) rather
95 than the normal 0 upto VL-1
96 * **VLi** VL inclusive: in fail-first mode, the truncation of
97 VL *includes* the current element at the failure point rather
98 than excludes it from the count.
100 ## Data-dependent fail-first on CR operations
102 The principle of data-dependent fail-first is that if, during the course
103 of sequentially evaluating an element's Condition Test, one such test
104 is encountered which fails, then VL (Vector Length) is truncated (set)
105 at that point. In the case of Arithmetic SVP64 Operations the Condition
106 Register Field generated from Rc=1 is used as the basis for the truncation
107 decision. However with CR-based operations that CR Field result to be
108 tested is provided *by the operation itself*.
110 Data-dependent SVP64 Vectorized Operations involving the creation
111 or modification of a CR can require an extra two bits, which are not
112 available in the compact space of the SVP64 RM `MODE` Field. With the
113 concept of element width overrides being meaningless for CR Fields it
114 is possible to use the `ELWIDTH` field for alternative purposes.
116 Condition Register based operations such as `sv.mfcr` and `sv.crand`
117 can thus be made more flexible. However the rules that apply in this
118 section also apply to future CR-based instructions.
120 There are two primary different types of CR operations:
122 * Those which have a 3-bit operand field (referring to a CR Field)
123 * Those which have a 5-bit operand (referring to a bit within the
126 Examining these two types it is observed that the difference may
127 be considered to be that the 5-bit variant *already* provides the
128 prerequisite information about which CR Field bit (EQ, GE, LT, SO) is
129 to be operated on by the instruction. Thus, logically, we may set the
132 * When a 5-bit CR Result field is used in an instruction, the
133 5-bit variant of Data-Dependent Fail-First
134 must be used. i.e. the bit of the CR field to be tested is
135 the one that has just been modified (created) by the operation.
136 * When a 3-bit CR Result field is used the 3-bit variant
137 must be used, providing as it does the missing `CRbit` field
138 in order to select which CR Field bit of the result shall
139 be tested (EQ, LE, GE, SO)
141 The reason why the 3-bit CR variant needs the additional CR-bit field
142 should be obvious from the fact that the 3-bit CR Field from the base
143 Power ISA v3.0B operation clearly does not contain and is missing the
144 two CR Field Selector bits. Thus, these two bits (to select EQ, LE,
145 GE or SO) must be provided in another way.
147 Examples of the former type:
149 * crand, cror, crnor. These all are 5-bit destination (BT). The bit
150 to be tested against `inv` is the one selected by `BT`
151 * mcrf. This has only 3-bit destiation (BF). In order to select the
152 bit to be tested, the alternative encoding must be used.
153 With `CRbit` coming from the SVP64 RM bits 22-23 the bit
154 of BF to be tested is identified.
156 Just as with SVP64 [[sv/branches]] there is the option to truncate
157 VL to include the element being tested (`VLi=1`) and to exclude it
160 Also exactly as with [[sv/normal]] fail-first, VL cannot, unlike
161 [[sv/ldst]], be set to an arbitrary value. Deterministic behaviour
164 **Apparent contradictory behaviour compared to Rc=1,VLi=0**
166 In [[openpower/sv/normal]] mode when Rc=1 and VLi=0 the Vector of
167 co-results appears to ignore VLi=0 because the last CR Field co-result
168 element tested is written out regardless of the setting of VLi.
169 This is because when Rc=1 the CR Fields are co-results *not* actual
172 When looking at the *actual* number of results written (arithmetic
173 results on arithmetic operations vs CR-Field results on *CR-Field*
174 operations), and ignoring the Rc=1 co-results entirely,
175 the totals (the behaviours) are consistent whether
178 *Programmer's Note: if necessary VL may be obtained by using the alias
179 `getvl`, followed by incrementing or decrementing the GPR with the
180 copy of VL, and using `setvl` with the same GPR, to adjust VL.
181 Programmers should be aware that MAXVL will always limit
184 ## Reduction and Iteration
186 Bearing in mind as described in the [[svp64/appendix]] SVP64 Horizontal
187 Reduction is a deterministic schedule on top of base Scalar v3.0
188 operations, the same rules apply to CR Operations, i.e. that programmers
189 must follow certain conventions in order for an *end result* of a
190 reduction to be achieved. Unlike other Vector ISAs *there are no explicit
191 reduction opcodes* in SVP64: Schedules however achieve the same effect.
193 Due to these conventions only reduction on operations such as `crand`
194 and `cror` are meaningful because these have Condition Register Fields
195 as both input and output. Meaningless operations are not prohibited
196 because the cost in hardware of doing so is prohibitive, but neither
197 are they `UNDEFINED`. Implementations are still required to execute them
198 but are at liberty to optimise out any operations that would ultimately
199 be overwritten, as long as Strict Program Order is still obvservable by
202 Also bear in mind that 'Reverse Gear' may be enabled, which can be
203 used in combination with overlapping CR operations to iteratively
204 accumulate results. Issuing a `sv.crand` operation for example with
205 `BA` differing from `BB` by one Condition Register Field would result
206 in a cascade effect, where the first-encountered CR Field would set the
207 result to zero, and also all subsequent CR Field elements thereafter:
210 # sv.crand/mr/rg CR4.ge.v, CR5.ge.v, CR4.ge.v
211 for i in VL-1 downto 0 # reverse gear
212 CR.field[4+i].ge &= CR.field[5+i].ge
215 `sv.crxor` with reduction would be particularly useful for parity
216 calculation for example, although there are many ways in which the same
217 calculation could be carried out (`parityw`)
218 after transferring a vector of CR Fields
219 to a GPR using crweird operations.
221 Implementations are free and clear to optimise these reductions in any way
222 they see fit, as long as the end-result is compatible with Strict Program
223 Order being observed, and Interrupt latency is not adversely impacted.
224 Good examples include `sv.cror/mr` which is a cumulative ORing of
225 a Vector of CR Field bits, and consequently an easy target for
228 ## Unusual and quirky CR operations
230 **cmp and other compare ops**
232 `cmp` and `cmpi` etc take GPRs as sources and create a CR Field as a result.
239 With `ELWIDTH` applying to the source GPR operands this is perfectly fine.
241 **crweird operations**
243 There are 4 weird CR-GPR operations and one reasonable one in
244 the [[cr_int_predication]] set:
250 * mcrfm - reasonably normal and referring to CR Fields for src and dest.
252 The "weird" operations have a non-standard behaviour, being able to
253 treat *individual bits* of a GPR effectively as elements. They are
254 expected to be Micro-coded by most Hardware implementations.
256 ## Effectively-separate Vector and Scalar Condition Register file
258 As mentioned in the introduction on [[sv/svp64]] some prohibitions
259 are made on instructions involving Condition Registers that allow
260 implementors to actually consider the Scalar CR (fields CR0-CR7)
261 as a completely separate register file from the Vector CRs
264 The complications arise for existing Hardware implementations
265 due to Power ISA not having had "Conditional Execution" added.
266 Adding entirely new pipelines and a new Vector CR Register file
267 is a much easier proposition to consider.
269 The prohibitions utilise the CR Field numbers implicitly to
270 split out Vectorized CR operations to be considered completely
271 separare and distinct from Scalar CR operations *even though
272 they both use the same binary encoding*. This does in turn
273 mean that at the Decode Phase it becomes necessary to examine
274 not only the operation (`sv.crand`, `sv.cmp`) but also
275 the CR Field numbers as well as whether, in the EXTRA2/3 Mode
276 bits, the operands are Vectorized.
278 A future version of Power ISA, where SVP64Single is proposed,
279 would in fact introduce "Conditional Execution", including
280 for VSX. At which point this prohibition becomes moot as
281 Predication would be required to be added into the existing
282 Scalar (and PackedSIMD VSX) side of existing Power ISA