static bool
loop_prefetch_arrays (struct loops *loops, struct loop *loop)
{
struct mem_ref_group *refs;
unsigned ahead, ninsns, unroll_factor;
struct tree_niter_desc desc;
bool unrolled = false;
/* Step 1: gather the memory references. */
refs = gather_memory_references (loop);
/* Step 2: estimate the reuse effects. */
prune_by_reuse (refs);
if (!anything_to_prefetch_p (refs))
goto fail;
/* Step 3: determine the ahead and unroll factor. */
/* FIXME: We should use not size of the loop, but the average number of
instructions executed per iteration of the loop. */
ninsns = tree_num_loop_insns (loop);
ahead = (PREFETCH_LATENCY + ninsns - 1) / ninsns;
unroll_factor = determine_unroll_factor (loop, refs, ahead, ninsns,
&desc);
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "Ahead %d, unroll factor %d\n", ahead, unroll_factor);
/* If the loop rolls less than the required unroll factor, prefetching
is useless. */
if (unroll_factor > 1
&& cst_and_fits_in_hwi (desc.niter)
&& (unsigned HOST_WIDE_INT) int_cst_value (desc.niter) < unroll_factor)
goto fail;
/* Step 4: what to prefetch? */
if (!schedule_prefetches (refs, unroll_factor, ahead))
goto fail;
/* Step 5: unroll the loop. TODO -- peeling of first and last few
iterations so that we do not issue superfluous prefetches. */
if (unroll_factor != 1)
{
tree_unroll_loop (loops, loop, unroll_factor,
single_dom_exit (loop), &desc);
unrolled = true;
}
/* Step 6: issue the prefetches. */
issue_prefetches (refs, unroll_factor, ahead);
fail:
release_mem_refs (refs);
return unrolled;
}
static bool
loop_prefetch_arrays (struct loop *loop)
{
struct mem_ref_group *refs;
unsigned ahead, ninsns, time, unroll_factor;
HOST_WIDE_INT est_niter;
struct tree_niter_desc desc;
bool unrolled = false, no_other_refs;
if (optimize_loop_nest_for_size_p (loop))
{
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, " ignored (cold area)\n");
return false;
}
/* Step 1: gather the memory references. */
refs = gather_memory_references (loop, &no_other_refs);
/* Step 2: estimate the reuse effects. */
prune_by_reuse (refs);
if (!anything_to_prefetch_p (refs))
goto fail;
determine_loop_nest_reuse (loop, refs, no_other_refs);
/* Step 3: determine the ahead and unroll factor. */
/* FIXME: the time should be weighted by the probabilities of the blocks in
the loop body. */
time = tree_num_loop_insns (loop, &eni_time_weights);
ahead = (PREFETCH_LATENCY + time - 1) / time;
est_niter = estimated_loop_iterations_int (loop, false);
/* The prefetches will run for AHEAD iterations of the original loop. Unless
the loop rolls at least AHEAD times, prefetching the references does not
make sense. */
if (est_niter >= 0 && est_niter <= (HOST_WIDE_INT) ahead)
{
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file,
"Not prefetching -- loop estimated to roll only %d times\n",
(int) est_niter);
goto fail;
}
mark_nontemporal_stores (loop, refs);
ninsns = tree_num_loop_insns (loop, &eni_size_weights);
unroll_factor = determine_unroll_factor (loop, refs, ninsns, &desc,
est_niter);
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "Ahead %d, unroll factor %d\n", ahead, unroll_factor);
/* Step 4: what to prefetch? */
if (!schedule_prefetches (refs, unroll_factor, ahead))
goto fail;
/* Step 5: unroll the loop. TODO -- peeling of first and last few
iterations so that we do not issue superfluous prefetches. */
if (unroll_factor != 1)
{
tree_unroll_loop (loop, unroll_factor,
single_dom_exit (loop), &desc);
unrolled = true;
}
/* Step 6: issue the prefetches. */
issue_prefetches (refs, unroll_factor, ahead);
fail:
release_mem_refs (refs);
return unrolled;
}
