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

[ARM64] Fix an issue where we were always assuming a copy was coming from a D subregister. 

llvm-svn: 207423
This commit is contained in:
Chad Rosier 2014-04-28 16:21:50 +00:00
parent 35897d97a3
commit 0def8e2652
2 changed files with 33 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
llvm/lib/Target/ARM64/ARM64AdvSIMDScalarPass.cpp
View File

@ -90,7 +90,7 @@ public:
virtual bool runOnMachineFunction(MachineFunction &F);
const char *getPassName() const {
return "AdvSIMD scalar operation optimization";
return "AdvSIMD Scalar Operation Optimization";
}
virtual void getAnalysisUsage(AnalysisUsage &AU) const {
@ -117,7 +117,7 @@ static bool isFPR64(unsigned Reg, unsigned SubReg,
SubReg == 0) ||
(MRI->getRegClass(Reg)->hasSuperClassEq(&ARM64::FPR128RegClass) &&
SubReg == ARM64::dsub);
// Physical register references just check the regist class directly.
// Physical register references just check the register class directly.
return (ARM64::FPR64RegClass.contains(Reg) && SubReg == 0) ||
(ARM64::FPR128RegClass.contains(Reg) && SubReg == ARM64::dsub);
}
@ -148,7 +148,7 @@ static unsigned getSrcFromCopy(const MachineInstr *MI,
MRI) &&
isFPR64(MI->getOperand(1).getReg(), MI->getOperand(1).getSubReg(),
MRI)) {
SubReg = ARM64::dsub;
SubReg = MI->getOperand(1).getSubReg();
return MI->getOperand(1).getReg();
}
}

31
llvm/test/CodeGen/ARM64/AdvSIMD-Scalar.ll
View File

@ -1,10 +1,15 @@
; RUN: llc < %s -march=arm64 -arm64-neon-syntax=apple -arm64-simd-scalar=true -asm-verbose=false | FileCheck %s
;
; RUN: llc < %s -march=arm64 -arm64-neon-syntax=generic -arm64-simd-scalar=true -asm-verbose=false | FileCheck %s -check-prefix=GENERIC
define <2 x i64> @bar(<2 x i64> %a, <2 x i64> %b) nounwind readnone {
; CHECK-LABEL: bar:
; CHECK: add.2d v[[REG:[0-9]+]], v0, v1
; CHECK: add d[[REG3:[0-9]+]], d[[REG]], d1
; CHECK: sub d[[REG2:[0-9]+]], d[[REG]], d1
; GENERIC-LABEL: bar:
; GENERIC: add v[[REG:[0-9]+]].2d, v0.2d, v1.2d
; GENERIC: add d[[REG3:[0-9]+]], d[[REG]], d1
; GENERIC: sub d[[REG2:[0-9]+]], d[[REG]], d1
%add = add <2 x i64> %a, %b
%vgetq_lane = extractelement <2 x i64> %add, i32 0
%vgetq_lane2 = extractelement <2 x i64> %b, i32 0
@ -19,6 +24,9 @@ define double @subdd_su64(<2 x i64> %a, <2 x i64> %b) nounwind readnone {
; CHECK-LABEL: subdd_su64:
; CHECK: sub d0, d1, d0
; CHECK-NEXT: ret
; GENERIC-LABEL: subdd_su64:
; GENERIC: sub d0, d1, d0
; GENERIC-NEXT: ret
%vecext = extractelement <2 x i64> %a, i32 0
%vecext1 = extractelement <2 x i64> %b, i32 0
%sub.i = sub nsw i64 %vecext1, %vecext
@ -30,9 +38,30 @@ define double @vaddd_su64(<2 x i64> %a, <2 x i64> %b) nounwind readnone {
; CHECK-LABEL: vaddd_su64:
; CHECK: add d0, d1, d0
; CHECK-NEXT: ret
; GENERIC-LABEL: vaddd_su64:
; GENERIC: add d0, d1, d0
; GENERIC-NEXT: ret
%vecext = extractelement <2 x i64> %a, i32 0
%vecext1 = extractelement <2 x i64> %b, i32 0
%add.i = add nsw i64 %vecext1, %vecext
%retval = bitcast i64 %add.i to double
ret double %retval
}
; sub MI doesn't access dsub register.
define double @add_sub_su64(<2 x i64> %a, <2 x i64> %b) nounwind readnone {
; CHECK-LABEL: add_sub_su64:
; CHECK: add d0, d1, d0
; CHECK: sub d0, {{d[0-9]+}}, d0
; CHECK-NEXT: ret
; GENERIC-LABEL: add_sub_su64:
; GENERIC: add d0, d1, d0
; GENERIC: sub d0, {{d[0-9]+}}, d0
; GENERIC-NEXT: ret
%vecext = extractelement <2 x i64> %a, i32 0
%vecext1 = extractelement <2 x i64> %b, i32 0
%add.i = add i64 %vecext1, %vecext
%sub.i = sub i64 0, %add.i
%retval = bitcast i64 %sub.i to double
ret double %retval
}