(no commit message)
[libreriscv.git] / openpower.mdwn
1 # OpenPOWER
2
3 In the late 1980s [[!wikipedia IBM]] developed a POWER family of processors.
4 This evolved to a specification known as the POWER ISA. In 2019 IBM made the POWER ISA [[!wikipedia Open_source]], to be looked after by the existing [[!wikipedia OpenPOWER_Foundation]]. Here is a longer history of [[!wikipedia IBM_POWER_microprocessors]]. These IBM proprietary processors
5 happen to implement what is now known as the POWER ISA. The names
6 POWER8, POWER9, POWER10 etc. are product designations equivalent to Intel
7 i5, i7, i9 etc. and are frequently conflated with versions of the POWER ISA (v2.07, v3.0c, v3.1b).
8
9 Libre-SOC is basing its [[Simple-V Vectorisation|sv]] CPU extensions on POWER ISA, because it wants to be able to specify a machine that can be completely trusted, and because POWER, thanks to IBM's involvement,
10 is designed for high performance.
11
12 See wikipedia page
13 <https://en.m.wikipedia.org/wiki/Power_ISA>
14
15 very useful resource describing all assembly instructions
16 <https://www.ibm.com/docs/en/aix/7.1?topic=reference-instruction-set>
17
18 # Evaluation
19
20 EULA released! looks good.
21 <https://openpowerfoundation.org/final-draft-of-the-power-isa-eula-released/>
22
23 # Links
24
25 * OpenPOWER Membership
26 <https://openpowerfoundation.org/join/>
27 * OpenPower HDL Mailing list <http://lists.mailinglist.openpowerfoundation.org/mailman/listinfo/openpower-hdl-cores>
28 * [[openpower/isatables]]
29 * [[openpower/whitepapers]]
30 * [[openpower/isa]] - pseudo-code extracted from POWER V3.0B PDF spec
31 * [[openpower/gem5]]
32 * [[openpower/sv]]
33 * [[openpower/opcode_regs_deduped]]
34 * [[openpower/simd_vsx]]
35 * [[openpower/ISA_WG]] - OpenPOWER ISA Working Group
36 * [[openpower/pearpc]]
37 * [[openpower/pipeline_operands]] - the allocation of operands on each pipeline
38 * [[3d_gpu/architecture/decoder]]
39 * <https://forums.raptorcs.com/>
40 * <http://lists.mailinglist.openpowerfoundation.org/mailman/listinfo/openpower-community-dev>
41 * <http://lists.mailinglist.openpowerfoundation.org/mailman/listinfo>
42 * <http://bugs.libre-riscv.org/show_bug.cgi?id=179>
43 * <https://openpowerfoundation.org/?resource_lib=power-isa-version-3-0>
44 * <https://openpowerfoundation.org/?resource_lib=ibm-power-isa-version-2-07-b>
45
46 PowerPC Unit Tests
47
48 * <https://github.com/lioncash/DolphinPPCTests>
49 * <https://github.com/JustinCB/macemu/blob/master/SheepShaver/src/kpx_cpu/src/test/test-powerpc.cpp>
50
51 Summary
52
53 * FP32 is converted to FP64. Requires SimpleV to be active.
54 * FP16 needed
55 * transcendental FP opcodes needed (sin, cos, atan2, root, log1p)
56 * FCVT between 16/32/64 needed
57 * c++11 atomics not very efficient
58 * no 16/48/64 opcodes, needs a shuffle of opcodes. TODO investigate Power VLE
59 * needs escape sequencing (ISAMUX/NS) - see [[openpower/isans_letter]]
60
61 # What we are *NOT* doing:
62
63 * A processor that is fundamentally incompatible (noncompliant) with Power.
64 (**escape-sequencing requires and guarantees compatibility**).
65 * Opcode 4 Signal Processing (SPE)
66 * Opcode 4 Vectors or Opcode 60 VSX (600+ additional instructions)
67 * Avoidable legacy opcodes
68 * SIMD. it's awful.
69
70 # SimpleV
71
72 see [[openpower/sv]].
73 SimpleV: a "hardware for-loop" which involves type-casting (both) the
74 register files to "a sequence of elements". The **one** instruction
75 (an unmodified **scalar** instruction) is interpreted as a *hardware
76 for-loop* that issues **multiple** internal instructions with
77 sequentially-incrementing register numbers.
78
79 Thus it is completely unnecessary to add any vector opcodes - at all -
80 saving hugely on both hardware and compiler development time when
81 the concept is dropped on top of a pre-existing ISA.
82
83 ## Condition Registers
84
85 Branch Facility (Section 2.3.1 V2.07B and V3.0B) has 4-bit registers: CR0 and CR1. When SimpleV is active, it may be better to set CR6 (the Vector CR field) instead.
86
87 ## Carry
88
89 SimpleV extends (wraps) *scalar* opcodes with a hardware-level for-loop. Therefore, each scalar operation with a carry-in and carry-out will **require its own carry in and out bit**. Most sensible location to use is the CRs
90
91 # Integer Overflow / Saturate
92
93 Typically used on vector operations (audio DSP), it makes no sense to have separate opcodes (Opcode 4 SPE). To be done instead as CSRs / vector-flags on *standard* arithmetic operations.
94
95 # atomics
96
97 Single instruction on RV, and x86, but multiple on Power. Needs investigation, particularly as to why cache flush exists.
98
99 https://www.cl.cam.ac.uk/~pes20/cpp/cpp0xmappings.html
100
101 Hot loops contain significant instruction count, really need new c++11 atomics. To be proposed as new extension because other OpenPower members will need them too
102
103 # FP16
104
105 Doesn't exist in Power, need to work out suitable opcodes, basically means duplicating the entire range of FP32/64 ops, symmetrically.
106
107 Usually done with a fmt field, 2 bit, last one is FP128
108
109 idea: rather than add dozens of new opcodes, add "repurposer" instructions that remap FP32 to 16/32/64/128 and FP64 likewise. can also be done as C instruction, only needs 4 bits to specify.
110
111 # Escape Sequencing
112
113 aka "ISAMUX/NS". Absolutely critical, also to have official endorsement
114 from OpenPower Foundation.
115
116 This will allow extending ISA (see ISAMUX/NS) in a clean fashion
117 (including for and by OpenPower Foundation)
118
119 ## Branches in namespaces
120
121 Branches are fine as it is up to the compiler to decide whether to let the
122 ISAMUX/NS/escape-sequence countdown run out.
123
124 This is all a software / compiler / ABI issue.
125
126 ## Function calls in namespaces
127
128 Storing and restoring the state of the page/subpage CSR should be done by the caller. Or, again, let the countdowns run out.
129
130 If certain alternative configs are expected, they are part of the function ABI which must be spec'd.
131
132 All of this is a software issue (compiler / ABI).
133
134 # Compressed, 48, 64, VBLOCK
135
136 TODO investigate Power VLE (Freescale doc Ref 314-68105)
137
138 Under Esc Seq, move mulli, twi, tdi out of major OP000 then use the
139 entire row, 2 bits instead of 3. greatly simplifies decoder.
140
141 * OP 000-000 and 000-001 for 16 bit compressed, 11 bit instructions
142 * OP 000-010 and 000-011 for 48 bit. 11 bits for SVP P48
143 * OP 000-100 and 000-201 for 64 bit. 11 bits for SVP P64
144 * OP 000-110 and 000-111 for VBLOCK. 11 bits available.
145
146 Note that this requires BE instruction encoding (separate from
147 data BE/LE encoding). BE encoding always places the major opcode in
148 the first 2 bytes of the raw (uninterpreted) sequential instruction
149 byte stream.
150
151 Thus in BE-instruction-mode, the first 2 bytes may be analysed to
152 detect whether the instruction is 16-bit Compressed, 48-bit SVP-P48,
153 64-bit SVP-64, variable-length VBLOCK, or plain 32-bit.
154
155 It is not possible to distinguish LE-encoded 32-bit instructions
156 from LE-encoded 16-bit instructions because in LE-encoded 32-bit
157 instructions, the opcode falls into:
158
159 * bytes 2 and 3 of any given raw (uninterpreted) sequential instruction
160 byte stream for a 32-bit instruction
161 * bytes 0 and 1 for a 16-bit Compressed instruction
162 * bytes 4 and 5 for a 48-bit SVP P48
163 * bytes 6 and 7 for a 64-bit SVP P64
164
165 Clearly this is an impossible situation, therefore BE is the only
166 option. Note: *this is completely separate from BE/LE for data*
167
168 # Compressed 16
169
170 Further "escape-sequencing".
171
172 Only 11 bits available. Idea: have "pages" where one instruction selects
173 the page number. It also specifies for how long that page is activated
174 (terminated on a branch)
175
176 The length to be a maximum of 4 bits, where 0b1111 indicates "permanently active".
177
178 Perhaps split OP000-000 and OP000-001 so that 2 pages can be active.
179
180 Store activation length in a CSR.
181
182 2nd idea: 11 bits can be used for extremely common operations, then length-encoding page selection for further ops, using the full 16 bit range and an entirely new encoding scheme. 1 bit specifies which of 2 pages was selected?
183
184 3rd idea: "stack" mechanism. Allow subpages like a stack, to page in new pages.
185
186 3 bits for subpage number. 4 bits for length, gives 7 bits. 4x7 is 28, then 3 bits can be used to specify "stack depth".
187
188 Requirements are to have one instruction in each subpage which resets all the way back to PowerISA default. The other is a "back up stack by 1".
189