Skip to content

[RISCV] Use Log2SEW=0 for VMNAND/VMSET created for riscv_vmsge(u) intrinsics. #119767

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.

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 4 commits into from
Dec 13, 2024
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
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
60 changes: 39 additions & 21 deletions llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -1664,32 +1664,50 @@ void RISCVDAGToDAGISel::Select(SDNode *Node) {
switch (RISCVTargetLowering::getLMUL(Src1VT)) {
default:
llvm_unreachable("Unexpected LMUL!");
#define CASE_VMSLT_VMNAND_VMSET_OPCODES(lmulenum, suffix, suffix_b) \
#define CASE_VMSLT_OPCODES(lmulenum, suffix) \
case RISCVII::VLMUL::lmulenum: \
VMSLTOpcode = IsUnsigned ? RISCV::PseudoVMSLTU_VX_##suffix \
: RISCV::PseudoVMSLT_VX_##suffix; \
VMSGTOpcode = IsUnsigned ? RISCV::PseudoVMSGTU_VX_##suffix \
: RISCV::PseudoVMSGT_VX_##suffix; \
VMNANDOpcode = RISCV::PseudoVMNAND_MM_##suffix_b; \
VMSetOpcode = RISCV::PseudoVMSET_M_##suffix_b; \
break;
CASE_VMSLT_VMNAND_VMSET_OPCODES(LMUL_F8, MF8, B64)
CASE_VMSLT_VMNAND_VMSET_OPCODES(LMUL_F4, MF4, B32)
CASE_VMSLT_VMNAND_VMSET_OPCODES(LMUL_F2, MF2, B16)
CASE_VMSLT_VMNAND_VMSET_OPCODES(LMUL_1, M1, B8)
CASE_VMSLT_VMNAND_VMSET_OPCODES(LMUL_2, M2, B4)
CASE_VMSLT_VMNAND_VMSET_OPCODES(LMUL_4, M4, B2)
CASE_VMSLT_VMNAND_VMSET_OPCODES(LMUL_8, M8, B1)
#undef CASE_VMSLT_VMNAND_VMSET_OPCODES
CASE_VMSLT_OPCODES(LMUL_F8, MF8)
CASE_VMSLT_OPCODES(LMUL_F4, MF4)
CASE_VMSLT_OPCODES(LMUL_F2, MF2)
CASE_VMSLT_OPCODES(LMUL_1, M1)
CASE_VMSLT_OPCODES(LMUL_2, M2)
CASE_VMSLT_OPCODES(LMUL_4, M4)
CASE_VMSLT_OPCODES(LMUL_8, M8)
#undef CASE_VMSLT_OPCODES
}
// Mask operations use the LMUL from the mask type.
switch (RISCVTargetLowering::getLMUL(VT)) {
default:
llvm_unreachable("Unexpected LMUL!");
#define CASE_VMNAND_VMSET_OPCODES(lmulenum, suffix) \
case RISCVII::VLMUL::lmulenum: \
VMNANDOpcode = RISCV::PseudoVMNAND_MM_##suffix; \
VMSetOpcode = RISCV::PseudoVMSET_M_##suffix; \
break;
CASE_VMNAND_VMSET_OPCODES(LMUL_F8, B64)
CASE_VMNAND_VMSET_OPCODES(LMUL_F4, B32)
CASE_VMNAND_VMSET_OPCODES(LMUL_F2, B16)
CASE_VMNAND_VMSET_OPCODES(LMUL_1, B8)
CASE_VMNAND_VMSET_OPCODES(LMUL_2, B4)
CASE_VMNAND_VMSET_OPCODES(LMUL_4, B2)
CASE_VMNAND_VMSET_OPCODES(LMUL_8, B1)
#undef CASE_VMNAND_VMSET_OPCODES
}
SDValue SEW = CurDAG->getTargetConstant(
Log2_32(Src1VT.getScalarSizeInBits()), DL, XLenVT);
SDValue MaskSEW = CurDAG->getTargetConstant(0, DL, XLenVT);
SDValue VL;
selectVLOp(Node->getOperand(3), VL);

// If vmsge(u) with minimum value, expand it to vmset.
if (IsCmpMinimum) {
ReplaceNode(Node, CurDAG->getMachineNode(VMSetOpcode, DL, VT, VL, SEW));
ReplaceNode(Node,
CurDAG->getMachineNode(VMSetOpcode, DL, VT, VL, MaskSEW));
return;
}

Expand All @@ -1708,7 +1726,7 @@ void RISCVDAGToDAGISel::Select(SDNode *Node) {
CurDAG->getMachineNode(VMSLTOpcode, DL, VT, {Src1, Src2, VL, SEW}),
0);
ReplaceNode(Node, CurDAG->getMachineNode(VMNANDOpcode, DL, VT,
{Cmp, Cmp, VL, SEW}));
{Cmp, Cmp, VL, MaskSEW}));
return;
}
case Intrinsic::riscv_vmsgeu_mask:
Expand Down Expand Up @@ -1742,7 +1760,7 @@ void RISCVDAGToDAGISel::Select(SDNode *Node) {
switch (RISCVTargetLowering::getLMUL(Src1VT)) {
default:
llvm_unreachable("Unexpected LMUL!");
#define CASE_VMSLT_OPCODES(lmulenum, suffix, suffix_b) \
#define CASE_VMSLT_OPCODES(lmulenum, suffix) \
case RISCVII::VLMUL::lmulenum: \
VMSLTOpcode = IsUnsigned ? RISCV::PseudoVMSLTU_VX_##suffix \
: RISCV::PseudoVMSLT_VX_##suffix; \
Expand All @@ -1751,13 +1769,13 @@ void RISCVDAGToDAGISel::Select(SDNode *Node) {
VMSGTMaskOpcode = IsUnsigned ? RISCV::PseudoVMSGTU_VX_##suffix##_MASK \
: RISCV::PseudoVMSGT_VX_##suffix##_MASK; \
break;
CASE_VMSLT_OPCODES(LMUL_F8, MF8, B64)
CASE_VMSLT_OPCODES(LMUL_F4, MF4, B32)
CASE_VMSLT_OPCODES(LMUL_F2, MF2, B16)
CASE_VMSLT_OPCODES(LMUL_1, M1, B8)
CASE_VMSLT_OPCODES(LMUL_2, M2, B4)
CASE_VMSLT_OPCODES(LMUL_4, M4, B2)
CASE_VMSLT_OPCODES(LMUL_8, M8, B1)
CASE_VMSLT_OPCODES(LMUL_F8, MF8)
CASE_VMSLT_OPCODES(LMUL_F4, MF4)
CASE_VMSLT_OPCODES(LMUL_F2, MF2)
CASE_VMSLT_OPCODES(LMUL_1, M1)
CASE_VMSLT_OPCODES(LMUL_2, M2)
CASE_VMSLT_OPCODES(LMUL_4, M4)
CASE_VMSLT_OPCODES(LMUL_8, M8)
#undef CASE_VMSLT_OPCODES
}
// Mask operations use the LMUL from the mask type.
Expand Down
2 changes: 1 addition & 1 deletion llvm/test/CodeGen/RISCV/rvv/vmsgeu.ll
Original file line number Diff line number Diff line change
Expand Up @@ -2183,7 +2183,7 @@ entry:
define <vscale x 4 x i1> @intrinsic_vmsgeu_vi_nxv4i16_i16(<vscale x 4 x i16> %0, iXLen %1) nounwind {
; CHECK-LABEL: intrinsic_vmsgeu_vi_nxv4i16_i16:
; CHECK: # %bb.0: # %entry
; CHECK-NEXT: vsetvli zero, a0, e16, m1, ta, ma
; CHECK-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
; CHECK-NEXT: vmset.m v0
; CHECK-NEXT: ret
entry:
Expand Down
Loading