RPCS3/llvm
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm.git synced 2025-04-05 06:41:51 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Revert "[FastISel][AArch64] Fold bit test and branch into TBZ and TBNZ."

Browse Source 
Reverting it until I have time to investigate a regression.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@218035 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Juergen Ributzka 2014-09-18 08:07:40 +00:00
parent ef48b51126
commit f789dac2dd
2 changed files with 8 additions and 82 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
lib/Target/AArch64/AArch64FastISel.cpp
View File

@ -1860,7 +1860,7 @@ static AArch64CC::CondCode getCompareCC(CmpInst::Predicate Pred) {
/// \brief Check if the comparison against zero and the following branch can be /// \brief Check if the comparison against zero and the following branch can be
/// folded into a single instruction (CBZ or CBNZ). /// folded into a single instruction (CBZ or CBNZ).
static bool canFoldZeroCheckIntoBranch(const CmpInst *CI) { static bool canFoldZeroIntoBranch(const CmpInst *CI) {
CmpInst::Predicate Predicate = CI->getPredicate(); CmpInst::Predicate Predicate = CI->getPredicate();
if ((Predicate != CmpInst::ICMP_EQ) && (Predicate != CmpInst::ICMP_NE)) if ((Predicate != CmpInst::ICMP_EQ) && (Predicate != CmpInst::ICMP_NE))
return false; return false;
@ -1918,7 +1918,7 @@ bool AArch64FastISel::selectBranch(const Instruction *I) {
} }
// Try to optimize comparisons against zero. // Try to optimize comparisons against zero.
if (canFoldZeroCheckIntoBranch(CI)) { if (canFoldZeroIntoBranch(CI)) {
const Value *LHS = CI->getOperand(0); const Value *LHS = CI->getOperand(0);
const Value *RHS = CI->getOperand(1); const Value *RHS = CI->getOperand(1);
@ -1927,33 +1927,12 @@ bool AArch64FastISel::selectBranch(const Instruction *I) {
if (C->isNullValue()) if (C->isNullValue())
std::swap(LHS, RHS); std::swap(LHS, RHS);
int TestBit = -1; static const unsigned OpcTable[2][2] = {
if (const auto *AI = dyn_cast<BinaryOperator>(LHS)) {AArch64::CBZW, AArch64::CBZX }, {AArch64::CBNZW, AArch64::CBNZX}
if (AI->getOpcode() == Instruction::And) {
const Value *AndLHS = AI->getOperand(0);
const Value *AndRHS = AI->getOperand(1);
if (const auto *C = dyn_cast<ConstantInt>(AndLHS))
if (C->getValue().isPowerOf2())
std::swap(AndLHS, AndRHS);
if (const auto *C = dyn_cast<ConstantInt>(AndRHS))
if (C->getValue().isPowerOf2()) {
TestBit = C->getValue().logBase2();
LHS = AndLHS;
}
}
static const unsigned OpcTable[2][2][2] = {
{ {AArch64::CBZW, AArch64::CBZX },
{AArch64::CBNZW, AArch64::CBNZX} },
{ {AArch64::TBZW, AArch64::TBZX },
{AArch64::TBNZW, AArch64::TBNZX} }
}; };
bool IsBitTest = TestBit != -1;
bool IsCmpNE = Predicate == CmpInst::ICMP_NE; bool IsCmpNE = Predicate == CmpInst::ICMP_NE;
bool Is64Bit = LHS->getType()->isIntegerTy(64); bool Is64Bit = LHS->getType()->isIntegerTy(64);
unsigned Opc = OpcTable[IsBitTest][IsCmpNE][Is64Bit]; unsigned Opc = OpcTable[IsCmpNE][Is64Bit];
unsigned SrcReg = getRegForValue(LHS); unsigned SrcReg = getRegForValue(LHS);
if (!SrcReg) if (!SrcReg)
@ -1961,12 +1940,9 @@ bool AArch64FastISel::selectBranch(const Instruction *I) {
bool SrcIsKill = hasTrivialKill(LHS); bool SrcIsKill = hasTrivialKill(LHS);
// Emit the combined compare and branch instruction. // Emit the combined compare and branch instruction.
MachineInstrBuilder MIB =
BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc)) BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc))
.addReg(SrcReg, getKillRegState(SrcIsKill)); .addReg(SrcReg, getKillRegState(SrcIsKill))
if (IsBitTest) .addMBB(TBB);
MIB.addImm(TestBit);
MIB.addMBB(TBB);
// Obtain the branch weight and add the TrueBB to the successor list. // Obtain the branch weight and add the TrueBB to the successor list.
uint32_t BranchWeight = 0; uint32_t BranchWeight = 0;

50
test/CodeGen/AArch64/fast-isel-tbz.ll
View File

@ -1,50 +0,0 @@
; RUN: llc -fast-isel -fast-isel-abort -aarch64-atomic-cfg-tidy=0 -verify-machineinstrs -mtriple=aarch64-apple-darwin < %s | FileCheck %s
define i32 @icmp_eq_i8(i8 zeroext %a) {
; CHECK-LABEL: icmp_eq_i8
; CHECK: tbz w0, #0, {{LBB.+_2}}
%1 = and i8 %a, 1
%2 = icmp eq i8 %1, 0
br i1 %2, label %bb1, label %bb2
bb2:
ret i32 1
bb1:
ret i32 0
}
define i32 @icmp_eq_i16(i16 zeroext %a) {
; CHECK-LABEL: icmp_eq_i16
; CHECK: tbz w0, #1, {{LBB.+_2}}
%1 = and i16 %a, 2
%2 = icmp eq i16 %1, 0
br i1 %2, label %bb1, label %bb2
bb2:
ret i32 1
bb1:
ret i32 0
}
define i32 @icmp_eq_i32(i32 %a) {
; CHECK-LABEL: icmp_eq_i32
; CHECK: tbz w0, #2, {{LBB.+_2}}
%1 = and i32 %a, 4
%2 = icmp eq i32 %1, 0
br i1 %2, label %bb1, label %bb2
bb2:
ret i32 1
bb1:
ret i32 0
}
define i32 @icmp_eq_i64(i64 %a) {
; CHECK-LABEL: icmp_eq_i64
; CHECK: tbz x0, #3, {{LBB.+_2}}
%1 = and i64 %a, 8
%2 = icmp eq i64 %1, 0
br i1 %2, label %bb1, label %bb2
bb2:
ret i32 1
bb1:
ret i32 0
}