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

[InstCombine] foldShiftOfShiftedLogic - add support for nonuniform constant vectors

This commit is contained in:
Simon Pilgrim 2020-10-09 14:22:19 +01:00
parent fe4715c47f
commit 9e796d5e71
2 changed files with 19 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
llvm/lib/Transforms/InstCombine/InstCombineShifts.cpp
View File

@ -328,8 +328,8 @@ static Instruction *foldShiftOfShiftedLogic(BinaryOperator &I,
if (!LogicInst || !LogicInst->isBitwiseLogicOp() || !LogicInst->hasOneUse()) if (!LogicInst || !LogicInst->isBitwiseLogicOp() || !LogicInst->hasOneUse())
return nullptr; return nullptr;
const APInt *C0, *C1; Constant *C0, *C1;
if (!match(I.getOperand(1), m_APInt(C1))) if (!match(I.getOperand(1), m_Constant(C1)))
return nullptr; return nullptr;
Instruction::BinaryOps ShiftOpcode = I.getOpcode(); Instruction::BinaryOps ShiftOpcode = I.getOpcode();
@ -341,9 +341,11 @@ static Instruction *foldShiftOfShiftedLogic(BinaryOperator &I,
Value *X, *Y; Value *X, *Y;
auto matchFirstShift = [&](Value *V) { auto matchFirstShift = [&](Value *V) {
BinaryOperator *BO; BinaryOperator *BO;
APInt Threshold(Ty->getScalarSizeInBits(), Ty->getScalarSizeInBits());
return match(V, m_BinOp(BO)) && BO->getOpcode() == ShiftOpcode && return match(V, m_BinOp(BO)) && BO->getOpcode() == ShiftOpcode &&
match(V, m_OneUse(m_Shift(m_Value(X), m_APInt(C0)))) && match(V, m_OneUse(m_Shift(m_Value(X), m_Constant(C0)))) &&
(*C0 + *C1).ult(Ty->getScalarSizeInBits()); match(ConstantExpr::getAdd(C0, C1),
m_SpecificInt_ICMP(ICmpInst::ICMP_ULT, Threshold));
}; };
// Logic ops are commutative, so check each operand for a match. // Logic ops are commutative, so check each operand for a match.
@ -355,7 +357,7 @@ static Instruction *foldShiftOfShiftedLogic(BinaryOperator &I,
return nullptr; return nullptr;
// shift (logic (shift X, C0), Y), C1 -> logic (shift X, C0+C1), (shift Y, C1) // shift (logic (shift X, C0), Y), C1 -> logic (shift X, C0+C1), (shift Y, C1)
Constant *ShiftSumC = ConstantInt::get(Ty, *C0 + *C1); Constant *ShiftSumC = ConstantExpr::getAdd(C0, C1);
Value *NewShift1 = Builder.CreateBinOp(ShiftOpcode, X, ShiftSumC); Value *NewShift1 = Builder.CreateBinOp(ShiftOpcode, X, ShiftSumC);
Value *NewShift2 = Builder.CreateBinOp(ShiftOpcode, Y, I.getOperand(1)); Value *NewShift2 = Builder.CreateBinOp(ShiftOpcode, Y, I.getOperand(1));
return BinaryOperator::Create(LogicInst->getOpcode(), NewShift1, NewShift2); return BinaryOperator::Create(LogicInst->getOpcode(), NewShift1, NewShift2);

24
llvm/test/Transforms/InstCombine/shift-logic.ll
View File

@ -16,9 +16,9 @@ define i8 @shl_and(i8 %x, i8 %y) {
define <2 x i8> @shl_and_nonuniform(<2 x i8> %x, <2 x i8> %y) { define <2 x i8> @shl_and_nonuniform(<2 x i8> %x, <2 x i8> %y) {
; CHECK-LABEL: @shl_and_nonuniform( ; CHECK-LABEL: @shl_and_nonuniform(
; CHECK-NEXT: [[SH0:%.*]] = shl <2 x i8> [[X:%.*]], <i8 3, i8 4> ; CHECK-NEXT: [[TMP1:%.*]] = shl <2 x i8> [[X:%.*]], <i8 5, i8 4>
; CHECK-NEXT: [[R:%.*]] = and <2 x i8> [[SH0]], [[Y:%.*]] ; CHECK-NEXT: [[TMP2:%.*]] = shl <2 x i8> [[Y:%.*]], <i8 2, i8 0>
; CHECK-NEXT: [[SH1:%.*]] = shl <2 x i8> [[R]], <i8 2, i8 0> ; CHECK-NEXT: [[SH1:%.*]] = and <2 x i8> [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret <2 x i8> [[SH1]] ; CHECK-NEXT: ret <2 x i8> [[SH1]]
; ;
%sh0 = shl <2 x i8> %x, <i8 3, i8 4> %sh0 = shl <2 x i8> %x, <i8 3, i8 4>
@ -45,9 +45,9 @@ define i16 @shl_or(i16 %x, i16 %py) {
define <2 x i16> @shl_or_undef(<2 x i16> %x, <2 x i16> %py) { define <2 x i16> @shl_or_undef(<2 x i16> %x, <2 x i16> %py) {
; CHECK-LABEL: @shl_or_undef( ; CHECK-LABEL: @shl_or_undef(
; CHECK-NEXT: [[Y:%.*]] = srem <2 x i16> [[PY:%.*]], <i16 42, i16 42> ; CHECK-NEXT: [[Y:%.*]] = srem <2 x i16> [[PY:%.*]], <i16 42, i16 42>
; CHECK-NEXT: [[SH0:%.*]] = shl <2 x i16> [[X:%.*]], <i16 5, i16 undef> ; CHECK-NEXT: [[TMP1:%.*]] = shl <2 x i16> [[X:%.*]], <i16 12, i16 undef>
; CHECK-NEXT: [[R:%.*]] = or <2 x i16> [[Y]], [[SH0]] ; CHECK-NEXT: [[TMP2:%.*]] = shl <2 x i16> [[Y]], <i16 7, i16 undef>
; CHECK-NEXT: [[SH1:%.*]] = shl <2 x i16> [[R]], <i16 7, i16 undef> ; CHECK-NEXT: [[SH1:%.*]] = or <2 x i16> [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret <2 x i16> [[SH1]] ; CHECK-NEXT: ret <2 x i16> [[SH1]]
; ;
%y = srem <2 x i16> %py, <i16 42, i16 42> ; thwart complexity-based canonicalization %y = srem <2 x i16> %py, <i16 42, i16 42> ; thwart complexity-based canonicalization
@ -72,9 +72,9 @@ define i32 @shl_xor(i32 %x, i32 %y) {
define <2 x i32> @shl_xor_nonuniform(<2 x i32> %x, <2 x i32> %y) { define <2 x i32> @shl_xor_nonuniform(<2 x i32> %x, <2 x i32> %y) {
; CHECK-LABEL: @shl_xor_nonuniform( ; CHECK-LABEL: @shl_xor_nonuniform(
; CHECK-NEXT: [[SH0:%.*]] = shl <2 x i32> [[X:%.*]], <i32 5, i32 6> ; CHECK-NEXT: [[TMP1:%.*]] = shl <2 x i32> [[X:%.*]], <i32 12, i32 14>
; CHECK-NEXT: [[R:%.*]] = xor <2 x i32> [[SH0]], [[Y:%.*]] ; CHECK-NEXT: [[TMP2:%.*]] = shl <2 x i32> [[Y:%.*]], <i32 7, i32 8>
; CHECK-NEXT: [[SH1:%.*]] = shl <2 x i32> [[R]], <i32 7, i32 8> ; CHECK-NEXT: [[SH1:%.*]] = xor <2 x i32> [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret <2 x i32> [[SH1]] ; CHECK-NEXT: ret <2 x i32> [[SH1]]
; ;
%sh0 = shl <2 x i32> %x, <i32 5, i32 6> %sh0 = shl <2 x i32> %x, <i32 5, i32 6>
@ -101,9 +101,9 @@ define i64 @lshr_and(i64 %x, i64 %py) {
define <2 x i64> @lshr_and_undef(<2 x i64> %x, <2 x i64> %py) { define <2 x i64> @lshr_and_undef(<2 x i64> %x, <2 x i64> %py) {
; CHECK-LABEL: @lshr_and_undef( ; CHECK-LABEL: @lshr_and_undef(
; CHECK-NEXT: [[Y:%.*]] = srem <2 x i64> [[PY:%.*]], <i64 42, i64 42> ; CHECK-NEXT: [[Y:%.*]] = srem <2 x i64> [[PY:%.*]], <i64 42, i64 42>
; CHECK-NEXT: [[SH0:%.*]] = lshr <2 x i64> [[X:%.*]], <i64 5, i64 undef> ; CHECK-NEXT: [[TMP1:%.*]] = lshr <2 x i64> [[X:%.*]], <i64 12, i64 undef>
; CHECK-NEXT: [[R:%.*]] = and <2 x i64> [[Y]], [[SH0]] ; CHECK-NEXT: [[TMP2:%.*]] = lshr <2 x i64> [[Y]], <i64 7, i64 undef>
; CHECK-NEXT: [[SH1:%.*]] = lshr <2 x i64> [[R]], <i64 7, i64 undef> ; CHECK-NEXT: [[SH1:%.*]] = and <2 x i64> [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret <2 x i64> [[SH1]] ; CHECK-NEXT: ret <2 x i64> [[SH1]]
; ;
%y = srem <2 x i64> %py, <i64 42, i64 42> ; thwart complexity-based canonicalization %y = srem <2 x i64> %py, <i64 42, i64 42> ; thwart complexity-based canonicalization