maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

ARM scheduling fix: compute predicated implicit use properly. 

Minor drive by fix to cleanup latency computation. Calling
getOperandLatency with a deliberately incorrect operand index does not
give you the latency you want.

llvm-svn: 158959
This commit is contained in:
Andrew Trick 2012-06-22 02:50:31 +00:00
parent 0d6989bd10
commit 3ccb1b8cf9
2 changed files with 4 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp
View File

@ -2998,9 +2998,7 @@ ARMBaseInstrInfo::getOutputLatency(const InstrItineraryData *ItinData,
return 1;
// If the second MI is predicated, then there is an implicit use dependency.
int Latency = getOperandLatency(ItinData, DefMI, DefIdx, DepMI,
DepMI->getNumOperands());
return (Latency <= 0) ? 1 : Latency;
return getInstrLatency(ItinData, DefMI);
}
unsigned ARMBaseInstrInfo::getInstrLatency(const InstrItineraryData *ItinData,

6
llvm/test/CodeGen/ARM/neon_div.ll
View File

@ -1,9 +1,9 @@
; RUN: llc < %s -march=arm -mattr=+neon -pre-RA-sched=source | FileCheck %s
; RUN: llc < %s -march=arm -mattr=+neon -pre-RA-sched=source -disable-post-ra | FileCheck %s
define <8 x i8> @sdivi8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
;CHECK: vrecpe.f32
;CHECK: vrecpe.f32
;CHECK: vmovn.i32
;CHECK: vrecpe.f32
;CHECK: vmovn.i32
;CHECK: vmovn.i16
%tmp1 = load <8 x i8>* %A
@ -15,10 +15,10 @@ define <8 x i8> @sdivi8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
define <8 x i8> @udivi8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
;CHECK: vrecpe.f32
;CHECK: vrecps.f32
;CHECK: vmovn.i32
;CHECK: vrecpe.f32
;CHECK: vrecps.f32
;CHECK: vmovn.i32
;CHECK: vmovn.i32
;CHECK: vqmovun.s16
%tmp1 = load <8 x i8>* %A
%tmp2 = load <8 x i8>* %B