https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79245
--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> --- (In reply to rguent...@suse.de from comment #5) > On Fri, 27 Jan 2017, jgreenhalgh at gcc dot gnu.org wrote: > > > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79245 > > > > --- Comment #4 from James Greenhalgh <jgreenhalgh at gcc dot gnu.org> --- > > (In reply to Richard Biener from comment #3) > > > Note the trivial fix will FAIL gcc.dg/tree-ssa/ldist-23.c which looks like > > > > > > int i; > > > for (i = 0; i < 128; ++i) > > > { > > > a[i] = a[i] + 1; > > > b[i] = d[i]; > > > c[i] = a[i] / d[i]; > > > } > > > > > > where the testcase expects b[i] = d[i] to be split out as memcpy but > > > the other two partitions to be fused. > > > > > > Generally the cost model lacks computing the number of input/output > > > streams > > > of a partition and a target interface to query it about limits. Usually > > > store bandwidth is not equal to load bandwidth and not re-used store > > > streams > > > can benefit from non-temporal stores being used by libc. > > > > > > In your testcase I wonder whether distributing to > > > > > > for (int j = 0; j < x; j++) > > > { > > > for (int i = 0; i < y; i++) > > > { > > > c[j][i] = b[j][i] - a[j][i]; > > > } > > > } > > > memcpy (a, b, ...); > > > > > > would be faster in the end (or even doing the memcpy first in this case). > > > > > > Well, for now let's be more conservative given the cost model really is > > > lacking. > > > > The testcase is reduced from CALC3 in 171.swim. I've been seeing a 3% > > regression for Cortex-A72 after r242038, and I can fix that with > > -fno-tree-loop-distribute-patterns. > > > > In that benchmark you've got 3 instances of the above pattern, so you end up > > with 3 memcpy calls after: > > > > DO 300 J=1,N > > DO 300 I=1,M > > UOLD(I,J) = U(I,J)+ALPHA*(UNEW(I,J)-2.*U(I,J)+UOLD(I,J)) > > VOLD(I,J) = V(I,J)+ALPHA*(VNEW(I,J)-2.*V(I,J)+VOLD(I,J)) > > POLD(I,J) = P(I,J)+ALPHA*(PNEW(I,J)-2.*P(I,J)+POLD(I,J)) > > U(I,J) = UNEW(I,J) > > V(I,J) = VNEW(I,J) > > P(I,J) = PNEW(I,J) > > 300 CONTINUE > > > > 3 memcpy calls compared to 3 vector store instructions doesn't seem the > > right > > tradeoff to me. Sorry if I reduced the testcase too far to make the balance > > clear. > > Itanic seems to like it though: > > http://gcc.opensuse.org/SPEC/CFP/sb-terbium-head-64/171_swim_big.png > > For Haswell I don't see any ups/downs for AMD Fam15 I see a slowdown > as well around that time. I guess it depends if the CPU is already > throttled by load/compute bandwith here. What should be profitable > is to distribute the above to three loops (w/o memcpy). So after > the patch doing -ftree-loop-distribution. Patch being > > Index: gcc/tree-loop-distribution.c > =================================================================== > --- gcc/tree-loop-distribution.c (revision 244963) > +++ gcc/tree-loop-distribution.c (working copy) > @@ -1548,8 +1548,7 @@ distribute_loop (struct loop *loop, vec< > for (int j = i + 1; > partitions.iterate (j, &partition); ++j) > { > - if (!partition_builtin_p (partition) > - && similar_memory_accesses (rdg, into, partition)) > + if (similar_memory_accesses (rdg, into, partition)) > { > if (dump_file && (dump_flags & TDF_DETAILS)) > { The distribution to three loops doesn't work because (fortran - eh) all arrays are part of the same (unnamed) common block and thus the memory accesses appear related -- in GIMPLE we see __BLNK__.uold[] and __BLNK__.unew[] and thus they appear as part of the same structure. The cost model is really too simple (not considering dependence distance or considering non-dependent accesses to be of infinite distance and thus unrelated - of course we compute dependence only after applying the cost model at the moment). It works with some extra (partly redundant) dependence checking: Index: gcc/tree-loop-distribution.c =================================================================== --- gcc/tree-loop-distribution.c (revision 244963) +++ gcc/tree-loop-distribution.c (working copy) @@ -1199,8 +1199,8 @@ ref_base_address (data_reference_p dr) accesses in RDG. */ static bool -similar_memory_accesses (struct graph *rdg, partition *partition1, - partition *partition2) +similar_memory_accesses (struct graph *rdg, vec<loop_p> loops, + partition *partition1, partition *partition2) { unsigned i, j, k, l; bitmap_iterator bi, bj; @@ -1228,7 +1228,18 @@ similar_memory_accesses (struct graph *r if (base1) FOR_EACH_VEC_ELT (RDG_DATAREFS (rdg, j), l, ref2) if (base1 == ref_base_address (ref2)) - return true; + { + ddr_p ddr = initialize_data_dependence_relation + (ref1, ref2, loops); + compute_affine_dependence (ddr, loops[0]); + tree res = DDR_ARE_DEPENDENT (ddr); + free_dependence_relation (ddr); + /* If the refs are independent then they constitute + different memory streams. + ??? Handle known large enough distance. */ + if (res != chrec_known) + return true; + } } } @@ -1548,8 +1559,7 @@ distribute_loop (struct loop *loop, vec< for (int j = i + 1; partitions.iterate (j, &partition); ++j) { - if (!partition_builtin_p (partition) - && similar_memory_accesses (rdg, into, partition)) + if (similar_memory_accesses (rdg, loop_nest, into, partition)) { if (dump_file && (dump_flags & TDF_DETAILS)) { I'm curious how that benchmarks for you (with -ftree-loop-distribution vs. without).
