openpower.mdwn

   1 # Evaluation
   2
   3 EULA released! looks good.
   4
   5 Links
   6
   7 * https://forums.raptorcs.com/
   8 * http://lists.mailinglist.openpowerfoundation.org/mailman/listinfo/openpower-community-dev
   9 * http://lists.mailinglist.openpowerfoundation.org/mailman/listinfo
  10 * http://bugs.libre-riscv.org/show_bug.cgi?id=179
  11
  12 Summary
  13
  14 * FP32 is converted to FP64. Requires SimpleV to be active.
  15 * FP16 needed
  16 * transcendental FP opcodes needed (sin, cos, atan2, root, log1p)
  17 * FCVT between 16/32/64 needed
  18 * c++11 atomics not very efficient
  19 * no 16/48/64 opcodes, needs a shuffle of opcodes.  TODO investigate Power VLE
  20 * needs escape sequencing (ISAMUX/NS)
  21
  22 # SimpleV
  23
  24 see [[simple_v_extension]] - will fit into 48/64/VBLOCK, see below.
  25 SimpleV: a "hardware for-loop" which involves type-casting (both) the
  26 register files to "a sequence of elements".  The **one** instruction
  27 (an unmodified **scalar** instruction) is interpreted as a *hardware
  28 for-loop* that issues **multiple** internal instructions with
  29 sequentially-incrementing register numbers.
  30
  31 Thus it is completely unnecessary to add any vector opcodes - at all -
  32 saving hugely on both hardware and compiler development time when
  33 the concept is dropped on top of a pre-existing ISA.
  34
  35 # atomics
  36
  37 Single instruction on RV, and x86, but multiple on Power. Needs investigation, particularly as to why cache flush exists.
  38
  39 https://www.cl.cam.ac.uk/~pes20/cpp/cpp0xmappings.html
  40
  41 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
  42
  43 # FP16
  44
  45 Doesn't exist in Power, need to work out suitable opcodes, basically means duplicating the entire range of FP32/64 ops, symmetrically.
  46
  47 Usually done with a fmt field, 2 bit, last one is FP128
  48
  49 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.
  50
  51 # Escape Sequencing
  52
  53 aka "ISAMUX/NS".  Absolutely critical, also to have official endorsement
  54 from OpenPower Foundation.
  55
  56 This will allow extending ISA (see ISAMUX/NS) in a clean fashion
  57 (including for and by OpenPower Foundation)
  58
  59 # Compressed, 48, 64, VBLOCK
  60
  61 TODO investigate Power VLE (Freescale doc Ref 314-68105)
  62
  63 Under Esc Seq, move mulli, twi, tdi out of major OP000 then use the
  64 entire row, 2 bits instead of 3.  greatly simplifies decoder.
  65
  66 * OP 000-000 and 000-001 for 16 bit compressed, 11 bit instructions
  67 * OP 000-010 and 000-011 for 48 bit. 11 bits for SVP P48
  68 * OP 000-100 and 000-201 for 64 bit. 11 bits for SVP P64
  69 * OP 000-110 and 000-111 for VBLOCK. 11 bits available.
  70
  71 # Compressed 16
  72
  73 Further "escape-sequencing".
  74
  75 Only 11 bits available. Idea: have "pages" where one instruction selects
  76 the page number. It also specifies for how long that page is activated
  77 (terminated on a branch)
  78
  79 The length to be a maximum of 4 bits, where 0b1111 indicates "permanently active".
  80
  81 Perhaps split OP000-000 and OP000-001 so that 2 pages can be active.
  82
  83 Store activation length in a CSR.
  84
  85 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?
  86
  87 3rd idea: "stack" mechanism. Allow subpages like a stack, to page in new pages.
  88
  89 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".
  90
  91 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".
  92
  93 # Branches
  94
  95 Branches are fine as it is up to the compiler to decide whether to let the
  96 ISAMUX/NS/escape-sequence countdown run out.
  97
  98 This is all a software / compiler / ABI issue.
  99
 100 # Function calls.
 101
 102 Storing and restoring the state of the page/subpage CSR should be done by the caller. Or, again, let the countdowns run out.
 103
 104 If certain alternative configs are expected, they are part of the function ABI which must be spec'd.
 105
 106 All of this is a software issue (compiler / ABI).
 107
 108 # RISCV userspace
 109
 110 Dual ISA, RV userspace only. Requires PowerISA to be able to context-switch RV registers and CSRs.
 111
 112 the exception entry point:
 113 https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/powerpc/kernel/exceptions-64s.S?h=v5.4-rc5#n409
 114
 115 the rest of the context switch code is in a different file:
 116 https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/powerpc/kernel/entry_64.S?h=v5.4-rc5#n589