6998f8ae2d
This patch simplifies the calculation of certain costs in getInstructionCost when isScalarAfterVectorization() returns a true value. There are a few places where we multiply a cost by a number N, i.e. unsigned N = isScalarAfterVectorization(I, VF) ? VF.getKnownMinValue() : 1; return N * TTI.getArithmeticInstrCost(... After some investigation it seems that there are only these cases that occur in practice: 1. VF is a scalar, in which case N = 1. 2. VF is a vector. We can only get here if: a) the instruction is a GEP/bitcast/PHI with scalar uses, or b) this is an update to an induction variable that remains scalar. I have changed the code so that N is assumed to always be 1. For GEPs the cost is always 0, since this is calculated later on as part of the load/store cost. PHI nodes are costed separately and were never previously multiplied by VF. For all other cases I have added an assert that none of the users needs scalarising, which didn't fire in any unit tests. Only one test required fixing and I believe the original cost for the scalar add instruction to have been wrong, since only one copy remains after vectorisation. I have also added a new test for the case when a pointer PHI feeds directly into a store that will be scalarised as we were previously never testing it. Differential Revision: https://reviews.llvm.org/D99718
41 lines
1.6 KiB
LLVM
41 lines
1.6 KiB
LLVM
; REQUIRES: asserts
|
|
; RUN: opt -loop-vectorize -force-vector-width=2 -debug-only=loop-vectorize -S -o - < %s 2>&1 | FileCheck %s
|
|
|
|
%struct.foo = type { i32, i64 }
|
|
|
|
; CHECK: LV: Found an estimated cost of 0 for VF 2 For instruction: %0 = bitcast i64* %b to i32*
|
|
|
|
; The bitcast below will be scalarized due to the predication in the loop. Bitcasts
|
|
; between pointer types should be treated as free, despite the scalarization.
|
|
define void @foo(%struct.foo* noalias nocapture %in, i32* noalias nocapture readnone %out, i64 %n) {
|
|
entry:
|
|
br label %for.body
|
|
|
|
for.body: ; preds = %entry, %if.end
|
|
%i.012 = phi i64 [ %inc, %if.end ], [ 0, %entry ]
|
|
%b = getelementptr inbounds %struct.foo, %struct.foo* %in, i64 %i.012, i32 1
|
|
%0 = bitcast i64* %b to i32*
|
|
%a = getelementptr inbounds %struct.foo, %struct.foo* %in, i64 %i.012, i32 0
|
|
%1 = load i32, i32* %a, align 8
|
|
%tobool.not = icmp eq i32 %1, 0
|
|
br i1 %tobool.not, label %if.end, label %land.lhs.true
|
|
|
|
land.lhs.true: ; preds = %for.body
|
|
%2 = load i32, i32* %0, align 4
|
|
%cmp2 = icmp sgt i32 %2, 0
|
|
br i1 %cmp2, label %if.then, label %if.end
|
|
|
|
if.then: ; preds = %land.lhs.true
|
|
%sub = add nsw i32 %2, -1
|
|
store i32 %sub, i32* %0, align 4
|
|
br label %if.end
|
|
|
|
if.end: ; preds = %if.then, %land.lhs.true, %for.body
|
|
%inc = add nuw nsw i64 %i.012, 1
|
|
%exitcond.not = icmp eq i64 %inc, %n
|
|
br i1 %exitcond.not, label %for.end, label %for.body
|
|
|
|
for.end: ; preds = %if.end
|
|
ret void
|
|
}
|