Hi,
To align with rs6000_insn_cost costing more for load type insns,
this patch is to make load insns cost more in vectorization cost
function. Considering that the result of load usually is used
somehow later (true-dep) but store won't, we keep the store as
before.
The SPEC2017 performance evaluation on Power8 shows 525.x264_r
+9.56%, 511.povray_r +2.08%, 527.cam4_r 1.16% gains, no
significant degradation, SPECINT geomean +0.88%, SPECFP geomean
+0.26%.
The SPEC2017 performance evaluation on Power9 shows no significant
improvement or degradation, SPECINT geomean +0.04%, SPECFP geomean
+0.04%.
The SPEC2006 performance evaluation on Power8 shows 454.calculix
+4.41% gain but 416.gamess -1.19% and 453.povray -3.83% degradation.
I looked into the two degradation bmks, the degradation were NOT
due to hotspot changes by vectorization, were all side effects.
SPECINT geomean +0.10%, SPECFP geomean no changed considering
the degradation.
Bootstrapped and regress tested on powerpc64le-linux-gnu.
Is OK for trunk?
BR,
Kewen
---
gcc/ChangeLog
2019-11-04 Kewen Lin <li...@gcc.gnu.org>
* config/rs6000/rs6000.c (rs6000_builtin_vectorization_cost): Make
scalar_load, vector_load, unaligned_load and vector_gather_load cost
a bit more on Power8 and up.
diff --git a/gcc/config/rs6000/rs6000.c b/gcc/config/rs6000/rs6000.c
index 5876714..876c7ef 100644
--- a/gcc/config/rs6000/rs6000.c
+++ b/gcc/config/rs6000/rs6000.c
@@ -4763,15 +4763,22 @@ rs6000_builtin_vectorization_cost (enum
vect_cost_for_stmt type_of_cost,
switch (type_of_cost)
{
case scalar_stmt:
- case scalar_load:
case scalar_store:
case vector_stmt:
- case vector_load:
case vector_store:
case vec_to_scalar:
case scalar_to_vec:
case cond_branch_not_taken:
return 1;
+ case scalar_load:
+ case vector_load:
+ /* Like rs6000_insn_cost, make load insns cost a bit more. FIXME: the
+ benefits were observed on Power8 and up, we can unify it if similar
+ profits are measured on Power6 and Power7. */
+ if (TARGET_P8_VECTOR)
+ return 2;
+ else
+ return 1;
case vec_perm:
/* Power7 has only one permute unit, make it a bit expensive. */
@@ -4792,8 +4799,9 @@ rs6000_builtin_vectorization_cost (enum
vect_cost_for_stmt type_of_cost,
case unaligned_load:
case vector_gather_load:
+ /* Like rs6000_insn_cost, make load insns cost a bit more. */
if (TARGET_EFFICIENT_UNALIGNED_VSX)
- return 1;
+ return 2;
if (TARGET_VSX && TARGET_ALLOW_MOVMISALIGN)
{
@@ -4827,7 +4835,13 @@ rs6000_builtin_vectorization_cost (enum
vect_cost_for_stmt type_of_cost,
/* Misaligned loads are not supported. */
gcc_unreachable ();
- return 2;
+ /* Like rs6000_insn_cost, make load insns cost a bit more. FIXME: the
+ benefits were observed on Power8 and up, we can unify it if similar
+ profits are measured on Power6 and Power7. */
+ if (TARGET_P8_VECTOR)
+ return 4;
+ else
+ return 2;
case unaligned_store:
case vector_scatter_store:
