CostModel: increase the default cost of supported floating point operations from 1 to two. Fixed a few tests that changes because now the cost of one insert + a vector operation on two doubles is lower than two scalar operations on doubles.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@179413 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Nadav Rotem 2013-04-12 21:15:03 +00:00
parent 196ee11f85
commit 9eb366acba
4 changed files with 38 additions and 10 deletions

View File

@ -204,20 +204,23 @@ unsigned BasicTTI::getArithmeticInstrCost(unsigned Opcode, Type *Ty,
std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Ty);
bool IsFloat = Ty->getScalarType()->isFloatingPointTy();
unsigned OpCost = (IsFloat ? 2 : 1);
if (TLI->isOperationLegalOrPromote(ISD, LT.second)) {
// The operation is legal. Assume it costs 1.
// If the type is split to multiple registers, assume that thre is some
// overhead to this.
// TODO: Once we have extract/insert subvector cost we need to use them.
if (LT.first > 1)
return LT.first * 2;
return LT.first * 1;
return LT.first * 2 * OpCost;
return LT.first * 1 * OpCost;
}
if (!TLI->isOperationExpand(ISD, LT.second)) {
// If the operation is custom lowered then assume
// thare the code is twice as expensive.
return LT.first * 2;
return LT.first * 2 * OpCost;
}
// Else, assume that we need to scalarize this op.
@ -230,7 +233,7 @@ unsigned BasicTTI::getArithmeticInstrCost(unsigned Opcode, Type *Ty,
}
// We don't know anything about this scalar instruction.
return 1;
return OpCost;
}
unsigned BasicTTI::getShuffleCost(ShuffleKind Kind, Type *Tp, int Index,

View File

@ -66,9 +66,9 @@ define void @avx2mull() {
; CHECK: fmul
define i32 @fmul(i32 %arg) {
;CHECK: cost of 1 {{.*}} fmul
;CHECK: cost of 2 {{.*}} fmul
%A = fmul <4 x float> undef, undef
;CHECK: cost of 1 {{.*}} fmul
;CHECK: cost of 2 {{.*}} fmul
%B = fmul <8 x float> undef, undef
ret i32 undef
}

View File

@ -34,7 +34,15 @@ for.body: ; preds = %for.body, %entry
%lftr.wideiv = trunc i64 %indvars.iv.next to i32
%exitcond = icmp eq i32 %lftr.wideiv, 10
br i1 %exitcond, label %for.end, label %for.body
; CHECK-NOT: <2 x double>
; CHECK: insertelement
; CHECK-NEXT: insertelement
; CHECK-NEXT: fadd <2 x double>
; CHECK-NEXT: insertelement
; CHECK-NEXT: insertelement
; CHECK-NEXT: fadd <2 x double>
; CHECK-NEXT: insertelement
; CHECK-NEXT: fmul <2 x double>
; CHECK-UNRL: %mul = fmul <2 x double> %2, %2
; CHECK-UNRL: %mul3 = fmul <2 x double> %2, %3
; CHECK-UNRL: %add = fadd <2 x double> %mul, %mul3

View File

@ -12,7 +12,11 @@ define double @test1(double %A1, double %A2, double %B1, double %B2) {
%R = fmul double %Z1, %Z2
ret double %R
; CHECK: @test1
; CHECK-NOT: fmul <2 x double>
; CHECK: fsub <2 x double>
; CHECK: fmul <2 x double>
; CHECK: fadd <2 x double>
; CHECK: extract
; CHECK: extract
; CHECK: ret double %R
}
@ -63,7 +67,12 @@ define double @test2(double %A1, double %A2, double %B1, double %B2) {
%R = fmul double %Z1, %Z2
ret double %R
; CHECK: @test2
; CHECK-NOT: fmul <2 x double>
; CHECK: insertelement
; CHECK: insertelement
; CHECK: insertelement
; CHECK: insertelement
; CHECK: fsub <2 x double>
; CHECK: fmul <2 x double>
; CHECK: ret double %R
}
@ -80,7 +89,15 @@ define double @test4(double %A1, double %A2, double %B1, double %B2) {
%R = fmul double %Z1, %Z2
ret double %R
; CHECK: @test4
; CHECK-NOT: fmul <2 x double>
; CHECK: insertelement
; CHECK: insertelement
; CHECK: insertelement
; CHECK: insertelement
; CHECK: fsub <2 x double>
; CHECK: fmul <2 x double>
; CHECK: insertelement
; CHECK: insertelement
; CHECK: fadd <2 x double>
; CHECK: ret double %R
}