gcc/doc/gcc/extensions-to-the-c-language-family/target-builtins/powerpc-matrix-multiply-assist-built-in-functions.rst

   1 ..
   2   Copyright 1988-2022 Free Software Foundation, Inc.
   3   This is part of the GCC manual.
   4   For copying conditions, see the copyright.rst file.
   5
   6 .. _powerpc-matrix-multiply-assist-built-in-functions:
   7
   8 PowerPC Matrix-Multiply Assist Built-in Functions
   9 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  10
  11 ISA 3.1 of the PowerPC added new Matrix-Multiply Assist (MMA) instructions.
  12 GCC provides support for these instructions through the following built-in
  13 functions which are enabled with the ``-mmma`` option.  The vec_t type
  14 below is defined to be a normal vector unsigned char type.  The uint2, uint4
  15 and uint8 parameters are 2-bit, 4-bit and 8-bit unsigned integer constants
  16 respectively.  The compiler will verify that they are constants and that
  17 their values are within range.
  18
  19 The built-in functions supported are:
  20
  21 .. code-block:: c++
  22
  23   void __builtin_mma_xvi4ger8 (__vector_quad *, vec_t, vec_t);
  24   void __builtin_mma_xvi8ger4 (__vector_quad *, vec_t, vec_t);
  25   void __builtin_mma_xvi16ger2 (__vector_quad *, vec_t, vec_t);
  26   void __builtin_mma_xvi16ger2s (__vector_quad *, vec_t, vec_t);
  27   void __builtin_mma_xvf16ger2 (__vector_quad *, vec_t, vec_t);
  28   void __builtin_mma_xvbf16ger2 (__vector_quad *, vec_t, vec_t);
  29   void __builtin_mma_xvf32ger (__vector_quad *, vec_t, vec_t);
  30
  31   void __builtin_mma_xvi4ger8pp (__vector_quad *, vec_t, vec_t);
  32   void __builtin_mma_xvi8ger4pp (__vector_quad *, vec_t, vec_t);
  33   void __builtin_mma_xvi8ger4spp(__vector_quad *, vec_t, vec_t);
  34   void __builtin_mma_xvi16ger2pp (__vector_quad *, vec_t, vec_t);
  35   void __builtin_mma_xvi16ger2spp (__vector_quad *, vec_t, vec_t);
  36   void __builtin_mma_xvf16ger2pp (__vector_quad *, vec_t, vec_t);
  37   void __builtin_mma_xvf16ger2pn (__vector_quad *, vec_t, vec_t);
  38   void __builtin_mma_xvf16ger2np (__vector_quad *, vec_t, vec_t);
  39   void __builtin_mma_xvf16ger2nn (__vector_quad *, vec_t, vec_t);
  40   void __builtin_mma_xvbf16ger2pp (__vector_quad *, vec_t, vec_t);
  41   void __builtin_mma_xvbf16ger2pn (__vector_quad *, vec_t, vec_t);
  42   void __builtin_mma_xvbf16ger2np (__vector_quad *, vec_t, vec_t);
  43   void __builtin_mma_xvbf16ger2nn (__vector_quad *, vec_t, vec_t);
  44   void __builtin_mma_xvf32gerpp (__vector_quad *, vec_t, vec_t);
  45   void __builtin_mma_xvf32gerpn (__vector_quad *, vec_t, vec_t);
  46   void __builtin_mma_xvf32gernp (__vector_quad *, vec_t, vec_t);
  47   void __builtin_mma_xvf32gernn (__vector_quad *, vec_t, vec_t);
  48
  49   void __builtin_mma_pmxvi4ger8 (__vector_quad *, vec_t, vec_t, uint4, uint4, uint8);
  50   void __builtin_mma_pmxvi4ger8pp (__vector_quad *, vec_t, vec_t, uint4, uint4, uint8);
  51
  52   void __builtin_mma_pmxvi8ger4 (__vector_quad *, vec_t, vec_t, uint4, uint4, uint4);
  53   void __builtin_mma_pmxvi8ger4pp (__vector_quad *, vec_t, vec_t, uint4, uint4, uint4);
  54   void __builtin_mma_pmxvi8ger4spp(__vector_quad *, vec_t, vec_t, uint4, uint4, uint4);
  55
  56   void __builtin_mma_pmxvi16ger2 (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2);
  57   void __builtin_mma_pmxvi16ger2s (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2);
  58   void __builtin_mma_pmxvf16ger2 (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2);
  59   void __builtin_mma_pmxvbf16ger2 (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2);
  60
  61   void __builtin_mma_pmxvi16ger2pp (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2);
  62   void __builtin_mma_pmxvi16ger2spp (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2);
  63   void __builtin_mma_pmxvf16ger2pp (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2);
  64   void __builtin_mma_pmxvf16ger2pn (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2);
  65   void __builtin_mma_pmxvf16ger2np (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2);
  66   void __builtin_mma_pmxvf16ger2nn (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2);
  67   void __builtin_mma_pmxvbf16ger2pp (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2);
  68   void __builtin_mma_pmxvbf16ger2pn (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2);
  69   void __builtin_mma_pmxvbf16ger2np (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2);
  70   void __builtin_mma_pmxvbf16ger2nn (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2);
  71
  72   void __builtin_mma_pmxvf32ger (__vector_quad *, vec_t, vec_t, uint4, uint4);
  73   void __builtin_mma_pmxvf32gerpp (__vector_quad *, vec_t, vec_t, uint4, uint4);
  74   void __builtin_mma_pmxvf32gerpn (__vector_quad *, vec_t, vec_t, uint4, uint4);
  75   void __builtin_mma_pmxvf32gernp (__vector_quad *, vec_t, vec_t, uint4, uint4);
  76   void __builtin_mma_pmxvf32gernn (__vector_quad *, vec_t, vec_t, uint4, uint4);
  77
  78   void __builtin_mma_xvf64ger (__vector_quad *, __vector_pair, vec_t);
  79   void __builtin_mma_xvf64gerpp (__vector_quad *, __vector_pair, vec_t);
  80   void __builtin_mma_xvf64gerpn (__vector_quad *, __vector_pair, vec_t);
  81   void __builtin_mma_xvf64gernp (__vector_quad *, __vector_pair, vec_t);
  82   void __builtin_mma_xvf64gernn (__vector_quad *, __vector_pair, vec_t);
  83
  84   void __builtin_mma_pmxvf64ger (__vector_quad *, __vector_pair, vec_t, uint4, uint2);
  85   void __builtin_mma_pmxvf64gerpp (__vector_quad *, __vector_pair, vec_t, uint4, uint2);
  86   void __builtin_mma_pmxvf64gerpn (__vector_quad *, __vector_pair, vec_t, uint4, uint2);
  87   void __builtin_mma_pmxvf64gernp (__vector_quad *, __vector_pair, vec_t, uint4, uint2);
  88   void __builtin_mma_pmxvf64gernn (__vector_quad *, __vector_pair, vec_t, uint4, uint2);
  89
  90   void __builtin_mma_xxmtacc (__vector_quad *);
  91   void __builtin_mma_xxmfacc (__vector_quad *);
  92   void __builtin_mma_xxsetaccz (__vector_quad *);
  93
  94   void __builtin_mma_build_acc (__vector_quad *, vec_t, vec_t, vec_t, vec_t);
  95   void __builtin_mma_disassemble_acc (void *, __vector_quad *);
  96
  97   void __builtin_vsx_build_pair (__vector_pair *, vec_t, vec_t);
  98   void __builtin_vsx_disassemble_pair (void *, __vector_pair *);
  99
 100   vec_t __builtin_vsx_xvcvspbf16 (vec_t);
 101   vec_t __builtin_vsx_xvcvbf16spn (vec_t);
 102
 103   __vector_pair __builtin_vsx_lxvp (size_t, __vector_pair *);
 104   void __builtin_vsx_stxvp (__vector_pair, size_t, __vector_pair *);