gcc/doc/gcc/gcc-command-options/machine-dependent-options/xtensa-options.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 .. program:: Xtensa
   7
   8 .. index:: Xtensa Options
   9
  10 .. _xtensa-options:
  11
  12 Xtensa Options
  13 ^^^^^^^^^^^^^^
  14
  15 These options are supported for Xtensa targets:
  16
  17 .. option:: -mconst16, -mno-const16
  18
  19   Enable or disable use of ``CONST16`` instructions for loading
  20   constant values.  The ``CONST16`` instruction is currently not a
  21   standard option from Tensilica.  When enabled, ``CONST16``
  22   instructions are always used in place of the standard ``L32R``
  23   instructions.  The use of ``CONST16`` is enabled by default only if
  24   the ``L32R`` instruction is not available.
  25
  26 .. option:: -mfused-madd, -mno-fused-madd
  27
  28   Enable or disable use of fused multiply/add and multiply/subtract
  29   instructions in the floating-point option.  This has no effect if the
  30   floating-point option is not also enabled.  Disabling fused multiply/add
  31   and multiply/subtract instructions forces the compiler to use separate
  32   instructions for the multiply and add/subtract operations.  This may be
  33   desirable in some cases where strict IEEE 754-compliant results are
  34   required: the fused multiply add/subtract instructions do not round the
  35   intermediate result, thereby producing results with *more* bits of
  36   precision than specified by the IEEE standard.  Disabling fused multiply
  37   add/subtract instructions also ensures that the program output is not
  38   sensitive to the compiler's ability to combine multiply and add/subtract
  39   operations.
  40
  41 .. option:: -mserialize-volatile, -mno-serialize-volatile
  42
  43   When this option is enabled, GCC inserts ``MEMW`` instructions before
  44   ``volatile`` memory references to guarantee sequential consistency.
  45   The default is :option:`-mserialize-volatile`.  Use
  46   :option:`-mno-serialize-volatile` to omit the ``MEMW`` instructions.
  47
  48 .. option:: -mforce-no-pic
  49
  50   For targets, like GNU/Linux, where all user-mode Xtensa code must be
  51   position-independent code (PIC), this option disables PIC for compiling
  52   kernel code.
  53
  54 .. option:: -mtext-section-literals, -mno-text-section-literals
  55
  56   These options control the treatment of literal pools.  The default is
  57   :option:`-mno-text-section-literals`, which places literals in a separate
  58   section in the output file.  This allows the literal pool to be placed
  59   in a data RAM/ROM, and it also allows the linker to combine literal
  60   pools from separate object files to remove redundant literals and
  61   improve code size.  With :option:`-mtext-section-literals`, the literals
  62   are interspersed in the text section in order to keep them as close as
  63   possible to their references.  This may be necessary for large assembly
  64   files.  Literals for each function are placed right before that function.
  65
  66 .. option:: -mauto-litpools, -mno-auto-litpools
  67
  68   These options control the treatment of literal pools.  The default is
  69   :option:`-mno-auto-litpools`, which places literals in a separate
  70   section in the output file unless :option:`-mtext-section-literals` is
  71   used.  With :option:`-mauto-litpools` the literals are interspersed in
  72   the text section by the assembler.  Compiler does not produce explicit
  73   ``.literal`` directives and loads literals into registers with
  74   ``MOVI`` instructions instead of ``L32R`` to let the assembler
  75   do relaxation and place literals as necessary.  This option allows
  76   assembler to create several literal pools per function and assemble
  77   very big functions, which may not be possible with
  78   :option:`-mtext-section-literals`.
  79
  80 .. option:: -mtarget-align, -mno-target-align
  81
  82   When this option is enabled, GCC instructs the assembler to
  83   automatically align instructions to reduce branch penalties at the
  84   expense of some code density.  The assembler attempts to widen density
  85   instructions to align branch targets and the instructions following call
  86   instructions.  If there are not enough preceding safe density
  87   instructions to align a target, no widening is performed.  The
  88   default is :option:`-mtarget-align`.  These options do not affect the
  89   treatment of auto-aligned instructions like ``LOOP``, which the
  90   assembler always aligns, either by widening density instructions or
  91   by inserting NOP instructions.
  92
  93 .. option:: -mlongcalls, -mno-longcalls
  94
  95   When this option is enabled, GCC instructs the assembler to translate
  96   direct calls to indirect calls unless it can determine that the target
  97   of a direct call is in the range allowed by the call instruction.  This
  98   translation typically occurs for calls to functions in other source
  99   files.  Specifically, the assembler translates a direct ``CALL``
 100   instruction into an ``L32R`` followed by a ``CALLX`` instruction.
 101   The default is :option:`-mno-longcalls`.  This option should be used in
 102   programs where the call target can potentially be out of range.  This
 103   option is implemented in the assembler, not the compiler, so the
 104   assembly code generated by GCC still shows direct call
 105   instructions---look at the disassembled object code to see the actual
 106   instructions.  Note that the assembler uses an indirect call for
 107   every cross-file call, not just those that really are out of range.
 108
 109 .. option:: -mabi={name}
 110
 111   Generate code for the specified ABI.  Permissible values are: :samp:`call0`,
 112   :samp:`windowed`.  Default ABI is chosen by the Xtensa core configuration.
 113
 114 .. option:: -mabi=call0
 115
 116   When this option is enabled function parameters are passed in registers
 117   ``a2`` through ``a7``, registers ``a12`` through ``a15`` are
 118   caller-saved, and register ``a15`` may be used as a frame pointer.
 119   When this version of the ABI is enabled the C preprocessor symbol
 120   ``__XTENSA_CALL0_ABI__`` is defined.
 121
 122 .. option:: -mabi=windowed
 123
 124   When this option is enabled function parameters are passed in registers
 125   ``a10`` through ``a15``, and called function rotates register window
 126   by 8 registers on entry so that its arguments are found in registers
 127   ``a2`` through ``a7``.  Register ``a7`` may be used as a frame
 128   pointer.  Register window is rotated 8 registers back upon return.
 129   When this version of the ABI is enabled the C preprocessor symbol
 130   ``__XTENSA_WINDOWED_ABI__`` is defined.
 131
 132 .. option:: -mextra-l32r-costs={n}
 133
 134   Specify an extra cost of instruction RAM/ROM access for ``L32R``
 135   instructions, in clock cycles.  This affects, when optimizing for speed,
 136   whether loading a constant from literal pool using ``L32R`` or
 137   synthesizing the constant from a small one with a couple of arithmetic
 138   instructions.  The default value is 0.