diff options
Diffstat (limited to 'gcc')
-rw-r--r-- | gcc/ChangeLog | 6 | ||||
-rw-r--r-- | gcc/doc/invoke.texi | 3 | ||||
-rw-r--r-- | gcc/tree-ssa-loop-prefetch.c | 29 |
3 files changed, 22 insertions, 16 deletions
diff --git a/gcc/ChangeLog b/gcc/ChangeLog index a0ba83a..8d01aa4 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,5 +1,11 @@ 2010-05-17 Changpeng Fang <changpeng.fang@amd.com> + * doc/invoke.texi: Update documentation for min-insn-to-prefetch-ratio. + * tree-ssa-loop-prefetch.c (is_loop_prefetching_profitable): Also apply + the insn to prefetch ratio heuristic to loops with known trip count. + +2010-05-17 Changpeng Fang <changpeng.fang@amd.com> + * tree-ssa-loop-prefetch.c (PREFETCH_MOD_TO_UNROLL_FACTOR_RATIO): New. (schedule_prefetches): Do not generate a prefetch if the unroll factor is far from what is required by the prefetch. diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi index e994902..bd29d24 100644 --- a/gcc/doc/invoke.texi +++ b/gcc/doc/invoke.texi @@ -8518,8 +8518,7 @@ The size of L2 cache, in kilobytes. @item min-insn-to-prefetch-ratio The minimum ratio between the number of instructions and the -number of prefetches to enable prefetching in a loop with an -unknown trip count. +number of prefetches to enable prefetching in a loop. @item prefetch-min-insn-to-mem-ratio The minimum ratio between the number of instructions and the diff --git a/gcc/tree-ssa-loop-prefetch.c b/gcc/tree-ssa-loop-prefetch.c index 4889604..4d85f54 100644 --- a/gcc/tree-ssa-loop-prefetch.c +++ b/gcc/tree-ssa-loop-prefetch.c @@ -1603,17 +1603,9 @@ is_loop_prefetching_profitable (unsigned ahead, HOST_WIDE_INT est_niter, return false; } - /* Profitability of prefetching is highly dependent on the trip count. - For a given AHEAD distance, the first AHEAD iterations do not benefit - from prefetching, and the last AHEAD iterations execute useless - prefetches. So, if the trip count is not large enough relative to AHEAD, - prefetching may cause serious performance degradation. To avoid this - problem when the trip count is not known at compile time, we - conservatively skip loops with high prefetching costs. For now, only - the I-cache cost is considered. The relative I-cache cost is estimated - by taking the ratio between the number of prefetches and the total - number of instructions. Since we are using integer arithmetic, we - compute the reciprocal of this ratio. + /* Prefetching most likely causes performance degradation when the instruction + to prefetch ratio is too small. Too many prefetch instructions in a loop + may reduce the I-cache performance. (unroll_factor * ninsns) is used to estimate the number of instructions in the unrolled loop. This implementation is a bit simplistic -- the number of issued prefetch instructions is also affected by unrolling. So, @@ -1623,12 +1615,21 @@ is_loop_prefetching_profitable (unsigned ahead, HOST_WIDE_INT est_niter, original loop * unroll_factor (at least the induction variable increases and the exit branches will get eliminated), so it might be better to use tree_estimate_loop_size + estimated_unrolled_size. */ - if (est_niter < 0) + insn_to_prefetch_ratio = (unroll_factor * ninsns) / prefetch_count; + if (insn_to_prefetch_ratio < MIN_INSN_TO_PREFETCH_RATIO) { - insn_to_prefetch_ratio = (unroll_factor * ninsns) / prefetch_count; - return insn_to_prefetch_ratio >= MIN_INSN_TO_PREFETCH_RATIO; + if (dump_file && (dump_flags & TDF_DETAILS)) + fprintf (dump_file, + "Not prefetching -- instruction to prefetch ratio (%d) too small\n", + insn_to_prefetch_ratio); + return false; } + /* Could not do further estimation if the trip count is unknown. Just assume + prefetching is profitable. Too aggressive??? */ + if (est_niter < 0) + return true; + if (est_niter < (HOST_WIDE_INT) (TRIP_COUNT_TO_AHEAD_RATIO * ahead)) { if (dump_file && (dump_flags & TDF_DETAILS)) |