Skip to content

[LoongArch] Custom legalize i32 operations for LA64 to reduce signed extensions #93811

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Jun 3, 2024
Merged

[LoongArch] Custom legalize i32 operations for LA64 to reduce signed extensions #93811

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
20 changes: 20 additions & 0 deletions llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -121,6 +121,8 @@ LoongArchTargetLowering::LoongArchTargetLowering(const TargetMachine &TM,
// Set operations for LA64 only.

if (Subtarget.is64Bit()) {
setOperationAction(ISD::ADD, MVT::i32, Custom);
setOperationAction(ISD::SUB, MVT::i32, Custom);
setOperationAction(ISD::SHL, MVT::i32, Custom);
setOperationAction(ISD::SRA, MVT::i32, Custom);
setOperationAction(ISD::SRL, MVT::i32, Custom);
Expand Down Expand Up @@ -1723,6 +1725,18 @@ static SDValue customLegalizeToWOp(SDNode *N, SelectionDAG &DAG, int NumOp,
return DAG.getNode(ISD::TRUNCATE, DL, N->getValueType(0), NewRes);
}

// Converts the given 32-bit operation to a i64 operation with signed extension
// semantic to reduce the signed extension instructions.
static SDValue customLegalizeToWOpWithSExt(SDNode *N, SelectionDAG &DAG) {
SDLoc DL(N);
SDValue NewOp0 = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(0));
SDValue NewOp1 = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(1));
SDValue NewWOp = DAG.getNode(N->getOpcode(), DL, MVT::i64, NewOp0, NewOp1);
SDValue NewRes = DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, MVT::i64, NewWOp,
DAG.getValueType(MVT::i32));
return DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, NewRes);
}

// Helper function that emits error message for intrinsics with/without chain
// and return a UNDEF or and the chain as the results.
static void emitErrorAndReplaceIntrinsicResults(
Expand Down Expand Up @@ -1846,6 +1860,12 @@ void LoongArchTargetLowering::ReplaceNodeResults(
switch (N->getOpcode()) {
default:
llvm_unreachable("Don't know how to legalize this operation");
case ISD::ADD:
case ISD::SUB:
assert(N->getValueType(0) == MVT::i32 && Subtarget.is64Bit() &&
"Unexpected custom legalisation");
Results.push_back(customLegalizeToWOpWithSExt(N, DAG));
break;
case ISD::UDIV:
case ISD::UREM:
assert(VT == MVT::i32 && Subtarget.is64Bit() &&
Expand Down
5 changes: 1 addition & 4 deletions llvm/lib/Target/LoongArch/LoongArchInstrInfo.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -441,7 +441,7 @@ def simm16_lsl16 : Operand<GRLenVT>,
// A 32-bit signed immediate expressible with a pair of `addu16i.d + addi` for
// use in additions.
def simm32_hi16_lo12: Operand<GRLenVT>, ImmLeaf<GRLenVT, [{
return isShiftedInt<16, 16>(Imm - SignExtend64<12>(Imm));
return !isInt<12>(Imm) && isShiftedInt<16, 16>(Imm - SignExtend64<12>(Imm));
}]>;

def BareSymbol : AsmOperandClass {
Expand Down Expand Up @@ -1106,11 +1106,8 @@ foreach Idx = 1...3 in {

let Predicates = [IsLA64] in {
def : PatGprGpr<add, ADD_D>;
def : PatGprGpr_32<add, ADD_W>;
def : PatGprImm<add, ADDI_D, simm12>;
def : PatGprImm_32<add, ADDI_W, simm12>;
def : PatGprGpr<sub, SUB_D>;
def : PatGprGpr_32<sub, SUB_W>;
def : PatGprGpr<sdiv, DIV_D>;
def : PatGprGpr_32<sdiv, DIV_W>;
def : PatGprGpr<udiv, DIV_DU>;
Expand Down
6 changes: 3 additions & 3 deletions llvm/test/CodeGen/LoongArch/alsl.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,7 +42,7 @@ define i32 @alsl_i32(i32 signext %a, i32 signext %b) nounwind {
;
; LA64-LABEL: alsl_i32:
; LA64: # %bb.0: # %entry
; LA64-NEXT: alsl.d $a0, $a0, $a1, 3
; LA64-NEXT: alsl.w $a0, $a0, $a1, 3
; LA64-NEXT: ret
entry:
%mul = mul nsw i32 %a, 8
Expand Down Expand Up @@ -178,7 +178,7 @@ define i32 @mul_add_i32(i32 signext %a, i32 signext %b) nounwind {
; LA64: # %bb.0: # %entry
; LA64-NEXT: slli.d $a2, $a0, 3
; LA64-NEXT: alsl.d $a0, $a0, $a2, 2
; LA64-NEXT: add.d $a0, $a1, $a0
; LA64-NEXT: add.w $a0, $a1, $a0
; LA64-NEXT: ret
entry:
%mul = mul nsw i32 %a, 12
Expand Down Expand Up @@ -325,7 +325,7 @@ define i32 @alsl_neg_i32(i32 signext %a, i32 signext %b) nounwind {
; LA64-LABEL: alsl_neg_i32:
; LA64: # %bb.0: # %entry
; LA64-NEXT: alsl.d $a0, $a0, $a0, 3
; LA64-NEXT: sub.d $a0, $a1, $a0
; LA64-NEXT: sub.w $a0, $a1, $a0
; LA64-NEXT: ret
entry:
%mul = mul nsw i32 %a, -9
Expand Down
4 changes: 2 additions & 2 deletions llvm/test/CodeGen/LoongArch/atomicrmw-uinc-udec-wrap.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -112,7 +112,7 @@ define i32 @atomicrmw_uinc_wrap_i32(ptr %ptr, i32 %val) {
; LA64-NEXT: # =>This Loop Header: Depth=1
; LA64-NEXT: # Child Loop BB2_3 Depth 2
; LA64-NEXT: move $a3, $a2
; LA64-NEXT: addi.d $a2, $a2, 1
; LA64-NEXT: addi.w $a2, $a2, 1
; LA64-NEXT: sltu $a4, $a3, $a1
; LA64-NEXT: xori $a4, $a4, 1
; LA64-NEXT: masknez $a4, $a2, $a4
Expand Down Expand Up @@ -298,7 +298,7 @@ define i32 @atomicrmw_udec_wrap_i32(ptr %ptr, i32 %val) {
; LA64-NEXT: # =>This Loop Header: Depth=1
; LA64-NEXT: # Child Loop BB6_3 Depth 2
; LA64-NEXT: move $a4, $a2
; LA64-NEXT: addi.d $a2, $a2, -1
; LA64-NEXT: addi.w $a2, $a2, -1
; LA64-NEXT: sltui $a5, $a4, 1
; LA64-NEXT: sltu $a6, $a3, $a4
; LA64-NEXT: masknez $a2, $a2, $a6
Expand Down
18 changes: 11 additions & 7 deletions llvm/test/CodeGen/LoongArch/ir-instruction/add.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,7 +54,7 @@ define i32 @add_i32(i32 %x, i32 %y) {
;
; LA64-LABEL: add_i32:
; LA64: # %bb.0:
; LA64-NEXT: add.d $a0, $a0, $a1
; LA64-NEXT: add.w $a0, $a0, $a1
; LA64-NEXT: ret
%add = add i32 %x, %y
ret i32 %add
Expand Down Expand Up @@ -144,7 +144,7 @@ define i32 @add_i32_3(i32 %x) {
;
; LA64-LABEL: add_i32_3:
; LA64: # %bb.0:
; LA64-NEXT: addi.d $a0, $a0, 3
; LA64-NEXT: addi.w $a0, $a0, 3
; LA64-NEXT: ret
%add = add i32 %x, 3
ret i32 %add
Expand Down Expand Up @@ -195,6 +195,7 @@ define i32 @add_i32_0x12340000(i32 %x) {
; LA64-LABEL: add_i32_0x12340000:
; LA64: # %bb.0:
; LA64-NEXT: addu16i.d $a0, $a0, 4660
; LA64-NEXT: addi.w $a0, $a0, 0
; LA64-NEXT: ret
%add = add i32 %x, 305397760
ret i32 %add
Expand Down Expand Up @@ -244,6 +245,7 @@ define i32 @add_i32_0x7fff0000(i32 %x) {
; LA64-LABEL: add_i32_0x7fff0000:
; LA64: # %bb.0:
; LA64-NEXT: addu16i.d $a0, $a0, 32767
; LA64-NEXT: addi.w $a0, $a0, 0
; LA64-NEXT: ret
%add = add i32 %x, 2147418112
ret i32 %add
Expand Down Expand Up @@ -293,6 +295,7 @@ define i32 @add_i32_minus_0x80000000(i32 %x) {
; LA64-LABEL: add_i32_minus_0x80000000:
; LA64: # %bb.0:
; LA64-NEXT: addu16i.d $a0, $a0, -32768
; LA64-NEXT: addi.w $a0, $a0, 0
; LA64-NEXT: ret
%add = add i32 %x, -2147483648
ret i32 %add
Expand Down Expand Up @@ -343,6 +346,7 @@ define i32 @add_i32_minus_0x10000(i32 %x) {
; LA64-LABEL: add_i32_minus_0x10000:
; LA64: # %bb.0:
; LA64-NEXT: addu16i.d $a0, $a0, -1
; LA64-NEXT: addi.w $a0, $a0, 0
; LA64-NEXT: ret
%add = add i32 %x, -65536
ret i32 %add
Expand Down Expand Up @@ -396,7 +400,7 @@ define i32 @add_i32_0x7fff07ff(i32 %x) {
; LA64-LABEL: add_i32_0x7fff07ff:
; LA64: # %bb.0:
; LA64-NEXT: addu16i.d $a0, $a0, 32767
; LA64-NEXT: addi.d $a0, $a0, 2047
; LA64-NEXT: addi.w $a0, $a0, 2047
; LA64-NEXT: ret
%add = add i32 %x, 2147420159
ret i32 %add
Expand Down Expand Up @@ -450,7 +454,7 @@ define i32 @add_i32_0x7ffef800(i32 %x) {
; LA64-LABEL: add_i32_0x7ffef800:
; LA64: # %bb.0:
; LA64-NEXT: addu16i.d $a0, $a0, 32767
; LA64-NEXT: addi.d $a0, $a0, -2048
; LA64-NEXT: addi.w $a0, $a0, -2048
; LA64-NEXT: ret
%add = add i32 %x, 2147416064
ret i32 %add
Expand Down Expand Up @@ -525,7 +529,7 @@ define i32 @add_i32_minus_0x23450679(i32 %x) {
; LA64-LABEL: add_i32_minus_0x23450679:
; LA64: # %bb.0:
; LA64-NEXT: addu16i.d $a0, $a0, -9029
; LA64-NEXT: addi.d $a0, $a0, -1657
; LA64-NEXT: addi.w $a0, $a0, -1657
; LA64-NEXT: ret
%add = add i32 %x, -591726201
ret i32 %add
Expand Down Expand Up @@ -580,7 +584,7 @@ define i32 @add_i32_minus_0x2345fedd(i32 %x) {
; LA64-LABEL: add_i32_minus_0x2345fedd:
; LA64: # %bb.0:
; LA64-NEXT: addu16i.d $a0, $a0, -9030
; LA64-NEXT: addi.d $a0, $a0, 291
; LA64-NEXT: addi.w $a0, $a0, 291
; LA64-NEXT: ret
%add = add i32 %x, -591789789
ret i32 %add
Expand Down Expand Up @@ -680,7 +684,7 @@ define i32 @add_i32_minus_0x80000800(i32 %x) {
; LA64: # %bb.0:
; LA64-NEXT: lu12i.w $a1, 524287
; LA64-NEXT: ori $a1, $a1, 2048
; LA64-NEXT: add.d $a0, $a0, $a1
; LA64-NEXT: add.w $a0, $a0, $a1
; LA64-NEXT: ret
%add = add i32 %x, -2147485696
ret i32 %add
Expand Down
2 changes: 1 addition & 1 deletion llvm/test/CodeGen/LoongArch/ir-instruction/and.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -444,7 +444,7 @@ define i32 @and_add_lsr(i32 %x, i32 %y) {
;
; LA64-LABEL: and_add_lsr:
; LA64: # %bb.0:
; LA64-NEXT: addi.d $a0, $a0, -1
; LA64-NEXT: addi.w $a0, $a0, -1
; LA64-NEXT: bstrpick.d $a1, $a1, 31, 20
; LA64-NEXT: and $a0, $a1, $a0
; LA64-NEXT: ret
Expand Down
2 changes: 1 addition & 1 deletion llvm/test/CodeGen/LoongArch/ir-instruction/call.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,7 +34,7 @@ define i32 @defined_function(i32 %a) nounwind {
;
; LA64-LABEL: defined_function:
; LA64: # %bb.0:
; LA64-NEXT: addi.d $a0, $a0, 1
; LA64-NEXT: addi.w $a0, $a0, 1
; LA64-NEXT: ret
%1 = add i32 %a, 1
ret i32 %1
Expand Down
4 changes: 2 additions & 2 deletions llvm/test/CodeGen/LoongArch/ir-instruction/load-store.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,7 @@ define i32 @load_store_global() nounwind {
; LA64NOPIC-NEXT: pcalau12i $a0, %pc_hi20(G)
; LA64NOPIC-NEXT: addi.d $a1, $a0, %pc_lo12(G)
; LA64NOPIC-NEXT: ld.w $a0, $a1, 0
; LA64NOPIC-NEXT: addi.d $a0, $a0, 1
; LA64NOPIC-NEXT: addi.w $a0, $a0, 1
; LA64NOPIC-NEXT: st.w $a0, $a1, 0
; LA64NOPIC-NEXT: ret
;
Expand All @@ -41,7 +41,7 @@ define i32 @load_store_global() nounwind {
; LA64PIC-NEXT: pcalau12i $a0, %pc_hi20(.LG$local)
; LA64PIC-NEXT: addi.d $a1, $a0, %pc_lo12(.LG$local)
; LA64PIC-NEXT: ld.w $a0, $a1, 0
; LA64PIC-NEXT: addi.d $a0, $a0, 1
; LA64PIC-NEXT: addi.w $a0, $a0, 1
; LA64PIC-NEXT: st.w $a0, $a1, 0
; LA64PIC-NEXT: ret
%v = load i32, ptr @G
Expand Down
2 changes: 1 addition & 1 deletion llvm/test/CodeGen/LoongArch/ir-instruction/sub.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,7 +54,7 @@ define i32 @sub_i32(i32 %x, i32 %y) {
;
; LA64-LABEL: sub_i32:
; LA64: # %bb.0:
; LA64-NEXT: sub.d $a0, $a0, $a1
; LA64-NEXT: sub.w $a0, $a0, $a1
; LA64-NEXT: ret
%sub = sub i32 %x, %y
ret i32 %sub
Expand Down
5 changes: 2 additions & 3 deletions llvm/test/CodeGen/LoongArch/select-to-shiftand.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -198,10 +198,9 @@ define i32 @sub_clamp_zero_i32(i32 signext %x, i32 signext %y) {
;
; LA64-LABEL: sub_clamp_zero_i32:
; LA64: # %bb.0:
; LA64-NEXT: sub.d $a2, $a0, $a1
; LA64-NEXT: sub.w $a0, $a0, $a1
; LA64-NEXT: srai.d $a0, $a0, 31
; LA64-NEXT: andn $a0, $a2, $a0
; LA64-NEXT: srai.d $a1, $a0, 31
; LA64-NEXT: andn $a0, $a0, $a1
; LA64-NEXT: ret
%sub = sub nsw i32 %x, %y
%cmp = icmp sgt i32 %sub, 0
Expand Down
29 changes: 19 additions & 10 deletions llvm/test/CodeGen/LoongArch/sextw-removal.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -793,7 +793,9 @@ define signext i32 @test14(i32 signext %0, i32 signext %1) {
; NORMV-NEXT: blt $a4, $a5, .LBB13_5
; NORMV-NEXT: # %bb.3: # in Loop: Header=BB13_2 Depth=1
; NORMV-NEXT: add.d $a0, $a3, $a0
; NORMV-NEXT: addi.w $a3, $a3, 1
; NORMV-NEXT: addi.d $a3, $a3, 1
; NORMV-NEXT: addi.w $a3, $a3, 0
; NORMV-NEXT: addi.d $a0, $a0, 0
; NORMV-NEXT: blt $a3, $a1, .LBB13_2
; NORMV-NEXT: .LBB13_4:
; NORMV-NEXT: addi.w $a0, $a0, 0
Expand Down Expand Up @@ -861,7 +863,9 @@ define signext i32 @test14b(i32 %0, i32 signext %1) {
; NORMV-NEXT: blt $a4, $a5, .LBB14_5
; NORMV-NEXT: # %bb.3: # in Loop: Header=BB14_2 Depth=1
; NORMV-NEXT: add.d $a0, $a3, $a0
; NORMV-NEXT: addi.w $a3, $a3, 1
; NORMV-NEXT: addi.d $a3, $a3, 1
; NORMV-NEXT: addi.w $a3, $a3, 0
; NORMV-NEXT: addi.d $a0, $a0, 0
; NORMV-NEXT: blt $a3, $a1, .LBB14_2
; NORMV-NEXT: .LBB14_4:
; NORMV-NEXT: addi.w $a0, $a0, 0
Expand Down Expand Up @@ -929,7 +933,9 @@ define signext i32 @test14c(i32 zeroext %0, i32 signext %1) {
; NORMV-NEXT: blt $a4, $a5, .LBB15_5
; NORMV-NEXT: # %bb.3: # in Loop: Header=BB15_2 Depth=1
; NORMV-NEXT: add.d $a0, $a3, $a0
; NORMV-NEXT: addi.w $a3, $a3, 1
; NORMV-NEXT: addi.d $a3, $a3, 1
; NORMV-NEXT: addi.w $a3, $a3, 0
; NORMV-NEXT: addi.d $a0, $a0, 0
; NORMV-NEXT: blt $a3, $a1, .LBB15_2
; NORMV-NEXT: .LBB15_4:
; NORMV-NEXT: addi.w $a0, $a0, 0
Expand Down Expand Up @@ -996,7 +1002,9 @@ define signext i32 @test14d(i31 zeroext %0, i32 signext %1) {
; NORMV-NEXT: blt $a4, $a5, .LBB16_5
; NORMV-NEXT: # %bb.3: # in Loop: Header=BB16_2 Depth=1
; NORMV-NEXT: add.d $a0, $a3, $a0
; NORMV-NEXT: addi.w $a3, $a3, 1
; NORMV-NEXT: addi.d $a3, $a3, 1
; NORMV-NEXT: addi.w $a3, $a3, 0
; NORMV-NEXT: addi.d $a0, $a0, 0
; NORMV-NEXT: blt $a3, $a1, .LBB16_2
; NORMV-NEXT: .LBB16_4:
; NORMV-NEXT: addi.w $a0, $a0, 0
Expand Down Expand Up @@ -1122,8 +1130,8 @@ define signext i32 @bug(i32 signext %x) {
; CHECK-NEXT: masknez $a1, $a1, $a2
; CHECK-NEXT: maskeqz $a2, $a4, $a2
; CHECK-NEXT: or $a1, $a2, $a1
; CHECK-NEXT: srai.d $a0, $a0, 31
; CHECK-NEXT: nor $a0, $a0, $zero
; CHECK-NEXT: srli.d $a0, $a0, 31
; CHECK-NEXT: add.w $a0, $a1, $a0
; CHECK-NEXT: ret
; CHECK-NEXT: .LBB18_2:
Expand Down Expand Up @@ -1176,13 +1184,13 @@ define signext i32 @bug(i32 signext %x) {
; NORMV-NEXT: masknez $a1, $a1, $a2
; NORMV-NEXT: maskeqz $a2, $a4, $a2
; NORMV-NEXT: or $a1, $a2, $a1
; NORMV-NEXT: srai.d $a0, $a0, 31
; NORMV-NEXT: nor $a0, $a0, $zero
; NORMV-NEXT: srli.d $a0, $a0, 31
; NORMV-NEXT: add.d $a0, $a1, $a0
; NORMV-NEXT: addi.w $a0, $a0, 0
; NORMV-NEXT: ret
; NORMV-NEXT: .LBB18_2:
; NORMV-NEXT: addi.w $a0, $zero, 0
; NORMV-NEXT: move $a0, $zero
; NORMV-NEXT: ret
entry:
%tobool.not = icmp eq i32 %x, 0
Expand Down Expand Up @@ -1345,7 +1353,7 @@ declare zeroext i16 @bat(i32 signext)
define signext i32 @sextw_sh2add(i1 zeroext %0, ptr %1, i32 signext %2, i32 signext %3, i32 signext %4) {
; CHECK-LABEL: sextw_sh2add:
; CHECK: # %bb.0:
; CHECK-NEXT: alsl.d $a2, $a2, $a3, 2
; CHECK-NEXT: alsl.w $a2, $a2, $a3, 2
; CHECK-NEXT: beqz $a0, .LBB21_2
; CHECK-NEXT: # %bb.1:
; CHECK-NEXT: st.w $a2, $a1, 0
Expand All @@ -1355,12 +1363,13 @@ define signext i32 @sextw_sh2add(i1 zeroext %0, ptr %1, i32 signext %2, i32 sign
;
; NORMV-LABEL: sextw_sh2add:
; NORMV: # %bb.0:
; NORMV-NEXT: alsl.d $a2, $a2, $a3, 2
; NORMV-NEXT: alsl.w $a2, $a2, $a3, 2
; NORMV-NEXT: beqz $a0, .LBB21_2
; NORMV-NEXT: # %bb.1:
; NORMV-NEXT: st.w $a2, $a1, 0
; NORMV-NEXT: .LBB21_2:
; NORMV-NEXT: add.w $a0, $a2, $a4
; NORMV-NEXT: add.d $a0, $a2, $a4
; NORMV-NEXT: addi.w $a0, $a0, 0
; NORMV-NEXT: ret
%6 = shl i32 %2, 2
%7 = add i32 %6, %3
Expand Down
Loading