Define PREFETCH_MOD_TO_UNROLL_FACTOR_RATIO heuristic.

[pf3gnuchains/gcc-fork.git] / gcc / tree-ssa-loop-prefetch.c

diff --git a/gcc/tree-ssa-loop-prefetch.c b/gcc/tree-ssa-loop-prefetch.c
index 2fc901d..4889604 100644 (file)
--- a/gcc/tree-ssa-loop-prefetch.c
+++ b/gcc/tree-ssa-loop-prefetch.c
@@ -225,6 +225,17 @@ struct mem_ref_group
 
 #define PREFETCH_ALL           (~(unsigned HOST_WIDE_INT) 0)
 
 
 #define PREFETCH_ALL           (~(unsigned HOST_WIDE_INT) 0)
 

+/* Do not generate a prefetch if the unroll factor is significantly less
+   than what is required by the prefetch.  This is to avoid redundant
+   prefetches.  For example, if prefetch_mod is 16 and unroll_factor is
+   1, this means prefetching requires unrolling the loop 16 times, but
+   the loop is not going to be unrolled.  In this case (ratio = 16),
+   prefetching is not likely to be beneficial.  */
+
+#ifndef PREFETCH_MOD_TO_UNROLL_FACTOR_RATIO
+#define PREFETCH_MOD_TO_UNROLL_FACTOR_RATIO 8
+#endif
+

 /* The memory reference.  */
 
 struct mem_ref
 /* The memory reference.  */
 
 struct mem_ref


@@ -716,6 +727,9 @@ prune_ref_by_group_reuse (struct mem_ref *ref, struct mem_ref *by,
       hit_from = ddown (delta_b, PREFETCH_BLOCK) * PREFETCH_BLOCK;
       prefetch_before = (hit_from - delta_r + step - 1) / step;
 
       hit_from = ddown (delta_b, PREFETCH_BLOCK) * PREFETCH_BLOCK;
       prefetch_before = (hit_from - delta_r + step - 1) / step;
 

+      /* Do not reduce prefetch_before if we meet beyond cache size.  */
+      if (prefetch_before > (unsigned) abs (L2_CACHE_SIZE_BYTES / step))
+        prefetch_before = PREFETCH_ALL;

       if (prefetch_before < ref->prefetch_before)
        ref->prefetch_before = prefetch_before;
 
       if (prefetch_before < ref->prefetch_before)
        ref->prefetch_before = prefetch_before;
 


@@ -746,6 +760,9 @@ prune_ref_by_group_reuse (struct mem_ref *ref, struct mem_ref *by,
                                reduced_prefetch_block, align_unit);
   if (miss_rate <= ACCEPTABLE_MISS_RATE)
     {
                                reduced_prefetch_block, align_unit);
   if (miss_rate <= ACCEPTABLE_MISS_RATE)
     {

+      /* Do not reduce prefetch_before if we meet beyond cache size.  */
+      if (prefetch_before > L2_CACHE_SIZE_BYTES / PREFETCH_BLOCK)
+        prefetch_before = PREFETCH_ALL;

       if (prefetch_before < ref->prefetch_before)
        ref->prefetch_before = prefetch_before;
 
       if (prefetch_before < ref->prefetch_before)
        ref->prefetch_before = prefetch_before;
 


@@ -915,6 +932,12 @@ schedule_prefetches (struct mem_ref_group *groups, unsigned unroll_factor,
        if (!should_issue_prefetch_p (ref))
          continue;
 
        if (!should_issue_prefetch_p (ref))
          continue;
 

+        /* The loop is far from being sufficiently unrolled for this
+           prefetch.  Do not generate prefetch to avoid many redudant
+           prefetches.  */
+        if (ref->prefetch_mod / unroll_factor > PREFETCH_MOD_TO_UNROLL_FACTOR_RATIO)
+          continue;
+

        /* If we need to prefetch the reference each PREFETCH_MOD iterations,
           and we unroll the loop UNROLL_FACTOR times, we need to insert
           ceil (UNROLL_FACTOR / PREFETCH_MOD) instructions in each
        /* If we need to prefetch the reference each PREFETCH_MOD iterations,
           and we unroll the loop UNROLL_FACTOR times, we need to insert
           ceil (UNROLL_FACTOR / PREFETCH_MOD) instructions in each