Re: [PATCH v2 02/13] regcache: Add functions suitable for regset_supply/collect.

[Andreas Arnez <arnez at linux dot vnet dot ibm dot com>]

On Mon, Jul 07 2014, Omair Javaid wrote:

> Is there a way around avoiding the loop in supply/collect where regnum
> != -1? It should be more efficient in cases where we are looking for a
> register with regnum > 0.

Good question.  The most straightforward way would be a register map
format where regnum is used as an index into an array of offsets, like
this:

int s390_regmap_gregset[S390_NUM_REGS] =
{
  /* Program Status Word.  */
  0x00, 0x04,
  /* General Purpose Registers.  */
  0x08, 0x0c, 0x10, 0x14,
  0x18, 0x1c, 0x20, 0x24,
  0x28, 0x2c, 0x30, 0x34,
  0x38, 0x3c, 0x40, 0x44,
  /* Access Registers.  */
  0x48, 0x4c, 0x50, 0x54,
  0x58, 0x5c, 0x60, 0x64,
  0x68, 0x6c, 0x70, 0x74,
  0x78, 0x7c, 0x80, 0x84,
  /* Floating Point Control Word.  */
  -1,
  /* Floating Point Registers.  */
  -1, -1, -1, -1, -1, -1, -1, -1,
  -1, -1, -1, -1, -1, -1, -1, -1,
  /* GPR Uppper Halves.  */
  -1, -1, -1, -1, -1, -1, -1, -1,
  -1, -1, -1, -1, -1, -1, -1, -1,
  /* GNU/Linux-specific optional "registers".  */
  0x88, -1, -1,
};

This is a real example.  For the full example refer to:

  https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;a=blob;f=gdb/s390-tdep.c;h=72d55450225e89da4394079efac1fa33b36cb68c;hb=d91fab15e7eb04f6c9b7fee5859d8815b7aa84ee#l412

As you see, this is where the s390 implementation came from.  Then I
realized that regnum > 0 is a very rare case, and that the common case
was suboptimal with this format, because the supply/collect functions
had to iterate over *all* registers, not just those of a specific
regset.

Patch #3 in this series expresses the regmap from above like this:

static const struct regcache_map_entry s390_gregmap[] =
{
  { 1, S390_PSWM_REGNUM },
  { 1, S390_PSWA_REGNUM },
  { 16, S390_R0_REGNUM },
  { 16, S390_A0_REGNUM },
  { 1, S390_ORIG_R2_REGNUM },
  { 0 }
};

In addition to being more efficient in the common case, I also consider
this version much easier to read and maintain.

We could certainly spend more effort on supplying and collecting a
single register more efficiently.  For instance, we could offer
additional routines for that special case, perhaps in conjunction with a
preparation function that converts a regmap to an indexed-by-regnum
array.  However, I wouldn't focus on that too much before actually
making use of it.  Note that currently these functions are *always*
called with regnum == -1.

In fact, it may be more adequate to completely get rid of the parameter
regnum in the regset supply/collect functions.  Any reason why we
shouldn't?