forked from OSchip/llvm-project
189900eb14
This stops reporting CostPerUse 1 for `R8`-`R15` and `XMM8`-`XMM31`.
This was previously done because instruction encoding require a REX
prefix when using them resulting in longer instruction encodings. I
found that this regresses the quality of the register allocation as the
costs impose an ordering on eviction candidates. I also feel that there
is a bit of an impedance mismatch as the actual costs occure when
encoding instructions using those registers, but the order of VReg
assignments is not primarily ordered by number of Defs+Uses.
I did extensive measurements with the llvm-test-suite wiht SPEC2006 +
SPEC2017 included, internal services showed similar patterns. Generally
there are a log of improvements but also a lot of regression. But on
average the allocation quality seems to improve at a small code size
regression.
Results for measuring static and dynamic instruction counts:
Dynamic Counts (scaled by execution frequency) / Optimization Remarks:
Spills+FoldedSpills -5.6%
Reloads+FoldedReloads -4.2%
Copies -0.1%
Static / LLVM Statistics:
regalloc.NumSpills mean -1.6%, geomean -2.8%
regalloc.NumReloads mean -1.7%, geomean -3.1%
size..text mean +0.4%, geomean +0.4%
Static / LLVM Statistics:
mean -2.2%, geomean -3.1%) regalloc.NumSpills
mean -2.6%, geomean -3.9%) regalloc.NumReloads
mean +0.6%, geomean +0.6%) size..text
Static / LLVM Statistics:
regalloc.NumSpills mean -3.0%
regalloc.NumReloads mean -3.3%
size..text mean +0.3%, geomean +0.3%
Differential Revision: https://reviews.llvm.org/D133902
|
||
|---|---|---|
| .. | ||
| AArch64 | ||
| AMDGPU | ||
| ARM | ||
| NVPTX | ||
| Power | ||
| RISCV | ||
| X86 | ||
| 2005-08-15-AddRecIV.ll | ||
| 2005-08-17-OutOfLoopVariant.ll | ||
| 2005-09-12-UsesOutOutsideOfLoop.ll | ||
| 2007-04-23-UseIterator.ll | ||
| 2008-08-13-CmpStride.ll | ||
| 2008-09-09-Overflow.ll | ||
| 2009-01-13-nonconstant-stride-outside-loop.ll | ||
| 2009-04-28-no-reduce-mul.ll | ||
| 2011-07-19-CritEdgeBreakCrash.ll | ||
| 2011-10-03-CritEdgeMerge.ll | ||
| 2011-10-06-ReusePhi.ll | ||
| 2011-10-13-SCEVChain.ll | ||
| 2011-10-14-IntPtr.ll | ||
| 2011-12-19-PostincQuadratic.ll | ||
| 2012-01-02-nopreheader.ll | ||
| 2012-01-16-nopreheader.ll | ||
| 2012-03-15-nopreheader.ll | ||
| 2012-03-26-constexpr.ll | ||
| 2012-07-13-ExpandUDiv.ll | ||
| 2012-07-18-LimitReassociate.ll | ||
| 2013-01-05-IndBr.ll | ||
| 2013-01-14-ReuseCast.ll | ||
| addrec-gep-address-space.ll | ||
| addrec-gep.ll | ||
| callbr-critical-edge-splitting.ll | ||
| callbr-critical-edge-splitting2.ll | ||
| canonical-form.ll | ||
| count-to-zero.ll | ||
| dbg-preserve-0.ll | ||
| dbg-preserve-1.ll | ||
| dbg-preserve-2.ll | ||
| dead-phi.ll | ||
| debuginfo-scev-salvage-0.ll | ||
| debuginfo-scev-salvage-1.ll | ||
| debuginfo-scev-salvage-2.ll | ||
| debuginfo-scev-salvage-3.ll | ||
| debuginfo-scev-salvage-4.ll | ||
| debuginfo-scev-salvage-5.ll | ||
| depth-limit-overrun.ll | ||
| different-type-ivs.ll | ||
| dominate-assert.ll | ||
| dont-hoist-simple-loop-constants.ll | ||
| dont_insert_redundant_ops.ll | ||
| dont_reduce_bytes.ll | ||
| dont_reverse.ll | ||
| ephemeral.ll | ||
| exit_compare_live_range.ll | ||
| funclet.ll | ||
| gnarly-setupcost.ll | ||
| hoist-parent-preheader.ll | ||
| illegal-addr-modes.ll | ||
| invariant_value_first.ll | ||
| invariant_value_first_arg.ll | ||
| ivchain.ll | ||
| lsr-comp-time.ll | ||
| lsr-overflow.ll | ||
| lsr-term-fold-negative-testcase.ll | ||
| lsr-term-fold.ll | ||
| missing-phi-operand-update.ll | ||
| multi-edge-latch.ll | ||
| negative-scale.ll | ||
| nested-reduce.ll | ||
| nonintegral.ll | ||
| nonlinear-postinc.ll | ||
| opaque-ptr.ll | ||
| ops_after_indvar.ll | ||
| optimizemax_debugloc.ll | ||
| phi_ehpad_ignore_sameval.ll | ||
| phi_node_update_multiple_preds.ll | ||
| post-inc-icmpzero.ll | ||
| post-inc-optsize.ll | ||
| post-increment-insertion.ll | ||
| pr2537.ll | ||
| pr2570.ll | ||
| pr3086.ll | ||
| pr3399.ll | ||
| pr3571.ll | ||
| pr12018.ll | ||
| pr12048.ll | ||
| pr12691.ll | ||
| pr18165.ll | ||
| pr25541.ll | ||
| pr27056.ll | ||
| pr31627.ll | ||
| pr48725.ll | ||
| pr50765.ll | ||
| pr50918.ll | ||
| pr51329.ll | ||
| pr51656.ll | ||
| pr52161.ll | ||
| pr56160.ll | ||
| preserve-gep-loop-variant.ll | ||
| related_indvars.ll | ||
| remove_indvar.ll | ||
| remove_scev_indvars.ll | ||
| scaling-factor-incompat-type.ll | ||
| scaling_factor_cost_crash.ll | ||
| scev-after-loopinstsimplify.ll | ||
| scev-expander-lcssa.ll | ||
| scev-insertpt-bug.ll | ||
| sext-ind-var.ll | ||
| share_code_in_preheader.ll | ||
| share_ivs.ll | ||
| shl.ll | ||
| two-combinations-bug.ll | ||
| uglygep-address-space.ll | ||
| uglygep.ll | ||
| use_postinc_value_outside_loop.ll | ||
| var_stride_used_by_compare.ll | ||
| variable_stride.ll | ||
| wrong-hoisting-iv.ll | ||