[llvm] r237947 - [PPC/LoopUnrollRuntime] Don't avoid high-cost trip count computation on the PPC/A2

[Hal Finkel]

Author: hfinkel
Date: Thu May 21 15:30:23 2015
New Revision: 237947

URL: http://llvm.org/viewvc/llvm-project?rev=237947&view=rev
Log:
[PPC/LoopUnrollRuntime] Don't avoid high-cost trip count computation on the PPC/A2

On X86 (and similar OOO cores) unrolling is very limited, and even if the
runtime unrolling is otherwise profitable, the expense of a division to compute
the trip count could greatly outweigh the benefits. On the A2, we unroll a lot,
and the benefits of unrolling are more significant (seeing a 5x or 6x speedup
is not uncommon), so we're more able to tolerate the expense, on average, of a
division to compute the trip count.

Added:
    llvm/trunk/test/Transforms/LoopUnroll/PowerPC/a2-high-cost-trip-count-computation.ll
Modified:
    llvm/trunk/lib/Target/PowerPC/PPCTargetTransformInfo.cpp

Modified: llvm/trunk/lib/Target/PowerPC/PPCTargetTransformInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PowerPC/PPCTargetTransformInfo.cpp?rev=237947&r1=237946&r2=237947&view=diff
==============================================================================
--- llvm/trunk/lib/Target/PowerPC/PPCTargetTransformInfo.cpp (original)
+++ llvm/trunk/lib/Target/PowerPC/PPCTargetTransformInfo.cpp Thu May 21 15:30:23 2015
@@ -187,6 +187,10 @@ void PPCTTIImpl::getUnrollingPreferences
     // The A2 is in-order with a deep pipeline, and concatenation unrolling
     // helps expose latency-hiding opportunities to the instruction scheduler.
     UP.Partial = UP.Runtime = true;
+
+    // We unroll a lot on the A2 (hundreds of instructions), and the benefits
+    // often outweigh the cost of a division to compute the trip count.
+    UP.AllowExpensiveTripCount = true;
   }
 
   BaseT::getUnrollingPreferences(L, UP);

Added: llvm/trunk/test/Transforms/LoopUnroll/PowerPC/a2-high-cost-trip-count-computation.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnroll/PowerPC/a2-high-cost-trip-count-computation.ll?rev=237947&view=auto
==============================================================================
--- llvm/trunk/test/Transforms/LoopUnroll/PowerPC/a2-high-cost-trip-count-computation.ll (added)
+++ llvm/trunk/test/Transforms/LoopUnroll/PowerPC/a2-high-cost-trip-count-computation.ll Thu May 21 15:30:23 2015
@@ -0,0 +1,27 @@
+; RUN: opt < %s -S -mtriple=powerpc64-unknown-linux-gnu -mcpu=a2 -loop-unroll | FileCheck %s
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+
+;; Check that we do emit expensive instructions to compute trip
+;; counts when unrolling loops on the a2 (because we unroll a lot).
+
+define i32 @test(i64 %v12, i8* %array, i64* %loc) {
+; CHECK-LABEL: @test(
+; CHECK: udiv
+entry:
+  %step = load i64, i64* %loc, !range !0
+  br label %loop
+
+loop:                                           ; preds = %entry, %loop
+  %k.015 = phi i64 [ %v15, %loop ], [ %v12, %entry ]
+  %v14 = getelementptr inbounds i8, i8* %array, i64 %k.015
+  store i8 0, i8* %v14
+  %v15 = add nuw nsw i64 %k.015, %step
+  %v16 = icmp slt i64 %v15, 8193
+  br i1 %v16, label %loop, label %loopexit
+
+loopexit:                             ; preds = %loop
+  ret i32 0
+}
+
+!0 = !{i64 1, i64 100}