[llvm] [RISCV][VLOPT] Add vector integer cmp instructions to isSupportedInstr and getOperandInfo (PR #119733)
Michael Maitland via llvm-commits
llvm-commits at lists.llvm.org
Thu Dec 12 10:10:59 PST 2024
https://github.com/michaelmaitland created https://github.com/llvm/llvm-project/pull/119733
None
>From ecb294e3814eeb4d7fafdd397fb9d4af69b8309f Mon Sep 17 00:00:00 2001
From: Michael Maitland <michaeltmaitland at gmail.com>
Date: Wed, 11 Dec 2024 10:50:13 -0800
Subject: [PATCH] [RISCV][VLOPT] Add vector integer cmp instructions to
isSupportedInstr and getOperandInfo
---
llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp | 47 +-
llvm/test/CodeGen/RISCV/rvv/vl-opt-instrs.ll | 419 ++++++++++++++++++
.../test/CodeGen/RISCV/rvv/vl-opt-op-info.mir | 81 +++-
3 files changed, 545 insertions(+), 2 deletions(-)
diff --git a/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp b/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp
index 1d5684d6038ea9..e3b21ec05171e8 100644
--- a/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp
+++ b/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp
@@ -521,6 +521,32 @@ static OperandInfo getOperandInfo(const MachineInstr &MI,
return OperandInfo(RISCVVType::getEMULEqualsEEWDivSEWTimesLMUL(0, MI), 0);
}
+ // Vector Integer Compare Instructions
+ // Dest EEW=1 and EMUL=(EEW/SEW)*LMUL. Source EEW=SEW and EMUL=LMUL.
+ case RISCV::VMSEQ_VI:
+ case RISCV::VMSEQ_VV:
+ case RISCV::VMSEQ_VX:
+ case RISCV::VMSNE_VI:
+ case RISCV::VMSNE_VV:
+ case RISCV::VMSNE_VX:
+ case RISCV::VMSLTU_VV:
+ case RISCV::VMSLTU_VX:
+ case RISCV::VMSLT_VV:
+ case RISCV::VMSLT_VX:
+ case RISCV::VMSLEU_VV:
+ case RISCV::VMSLEU_VI:
+ case RISCV::VMSLEU_VX:
+ case RISCV::VMSLE_VV:
+ case RISCV::VMSLE_VI:
+ case RISCV::VMSLE_VX:
+ case RISCV::VMSGTU_VI:
+ case RISCV::VMSGTU_VX:
+ case RISCV::VMSGT_VI:
+ case RISCV::VMSGT_VX:
+ if (IsMODef)
+ return OperandInfo(RISCVVType::getEMULEqualsEEWDivSEWTimesLMUL(0, MI), 0);
+ return OperandInfo(MIVLMul, MILog2SEW);
+
default:
return {};
}
@@ -599,7 +625,26 @@ static bool isSupportedInstr(const MachineInstr &MI) {
case RISCV::VNSRA_WV:
case RISCV::VNSRA_WX:
// Vector Integer Compare Instructions
- // FIXME: Add support
+ case RISCV::VMSEQ_VI:
+ case RISCV::VMSEQ_VV:
+ case RISCV::VMSEQ_VX:
+ case RISCV::VMSNE_VI:
+ case RISCV::VMSNE_VV:
+ case RISCV::VMSNE_VX:
+ case RISCV::VMSLTU_VV:
+ case RISCV::VMSLTU_VX:
+ case RISCV::VMSLT_VV:
+ case RISCV::VMSLT_VX:
+ case RISCV::VMSLEU_VV:
+ case RISCV::VMSLEU_VI:
+ case RISCV::VMSLEU_VX:
+ case RISCV::VMSLE_VV:
+ case RISCV::VMSLE_VI:
+ case RISCV::VMSLE_VX:
+ case RISCV::VMSGTU_VI:
+ case RISCV::VMSGTU_VX:
+ case RISCV::VMSGT_VI:
+ case RISCV::VMSGT_VX:
// Vector Integer Min/Max Instructions
case RISCV::VMINU_VV:
case RISCV::VMINU_VX:
diff --git a/llvm/test/CodeGen/RISCV/rvv/vl-opt-instrs.ll b/llvm/test/CodeGen/RISCV/rvv/vl-opt-instrs.ll
index a21e3df85193fb..eda78733429024 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vl-opt-instrs.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vl-opt-instrs.ll
@@ -1082,6 +1082,425 @@ define <vscale x 4 x i16> @vnsra_wv(<vscale x 4 x i32> %a, <vscale x 4 x i16> %b
ret <vscale x 4 x i16> %2
}
+define <vscale x 4 x i1> @vmseq_vi(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, iXLen %vl) {
+; NOVLOPT-LABEL: vmseq_vi:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmseq.vi v10, v8, 5
+; NOVLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v10, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmseq_vi:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmseq.vi v10, v8, 5
+; VLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v10, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmseq.nxv4i32.i32(<vscale x 4 x i32> %a, i32 5, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
+
+define <vscale x 4 x i1> @vmseq_vx(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, i32 %c, iXLen %vl) {
+; NOVLOPT-LABEL: vmseq_vx:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmseq.vx v10, v8, a0
+; NOVLOPT-NEXT: vsetvli zero, a1, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v10, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmseq_vx:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmseq.vx v10, v8, a0
+; VLOPT-NEXT: vsetvli zero, a1, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v10, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmseq.nxv4i32.i32(<vscale x 4 x i32> %a, i32 %c, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
+
+define <vscale x 4 x i1> @vmseq_vv(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, <vscale x 4 x i32> %c, iXLen %vl) {
+; NOVLOPT-LABEL: vmseq_vv:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmseq.vv v12, v8, v10
+; NOVLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v12, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmseq_vv:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmseq.vv v12, v8, v10
+; VLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v12, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmseq.nxv4i32.nxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %c, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
+
+define <vscale x 4 x i1> @vmsne_vi(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, iXLen %vl) {
+; NOVLOPT-LABEL: vmsne_vi:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmsne.vi v10, v8, 5
+; NOVLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v10, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmsne_vi:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmsne.vi v10, v8, 5
+; VLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v10, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmsne.nxv4i32.i32(<vscale x 4 x i32> %a, i32 5, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
+
+define <vscale x 4 x i1> @vmsne_vx(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, i32 %c, iXLen %vl) {
+; NOVLOPT-LABEL: vmsne_vx:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmsne.vx v10, v8, a0
+; NOVLOPT-NEXT: vsetvli zero, a1, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v10, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmsne_vx:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmsne.vx v10, v8, a0
+; VLOPT-NEXT: vsetvli zero, a1, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v10, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmsne.nxv4i32.i32(<vscale x 4 x i32> %a, i32 %c, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
+
+define <vscale x 4 x i1> @vmsne_vv(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, <vscale x 4 x i32> %c, iXLen %vl) {
+; NOVLOPT-LABEL: vmsne_vv:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmsne.vv v12, v8, v10
+; NOVLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v12, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmsne_vv:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmsne.vv v12, v8, v10
+; VLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v12, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmsne.nxv4i32.nxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %c, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
+
+define <vscale x 4 x i1> @vmsltu_vx(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, i32 %c, iXLen %vl) {
+; NOVLOPT-LABEL: vmsltu_vx:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmsltu.vx v10, v8, a0
+; NOVLOPT-NEXT: vsetvli zero, a1, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v10, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmsltu_vx:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmsltu.vx v10, v8, a0
+; VLOPT-NEXT: vsetvli zero, a1, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v10, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmsltu.nxv4i32.i32(<vscale x 4 x i32> %a, i32 %c, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
+
+define <vscale x 4 x i1> @vmsltu_vv(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, <vscale x 4 x i32> %c, iXLen %vl) {
+; NOVLOPT-LABEL: vmsltu_vv:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmsltu.vv v12, v8, v10
+; NOVLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v12, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmsltu_vv:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmsltu.vv v12, v8, v10
+; VLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v12, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmsltu.nxv4i32.nxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %c, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
+
+define <vscale x 4 x i1> @vmslt_vx(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, i32 %c, iXLen %vl) {
+; NOVLOPT-LABEL: vmslt_vx:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmslt.vx v10, v8, a0
+; NOVLOPT-NEXT: vsetvli zero, a1, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v10, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmslt_vx:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmslt.vx v10, v8, a0
+; VLOPT-NEXT: vsetvli zero, a1, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v10, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmslt.nxv4i32.i32(<vscale x 4 x i32> %a, i32 %c, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
+
+define <vscale x 4 x i1> @vmslt_vv(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, <vscale x 4 x i32> %c, iXLen %vl) {
+; NOVLOPT-LABEL: vmslt_vv:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmslt.vv v12, v8, v10
+; NOVLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v12, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmslt_vv:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmslt.vv v12, v8, v10
+; VLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v12, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmslt.nxv4i32.nxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %c, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
+
+define <vscale x 4 x i1> @vmsleu_vi(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, iXLen %vl) {
+; NOVLOPT-LABEL: vmsleu_vi:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmsleu.vi v10, v8, 5
+; NOVLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v10, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmsleu_vi:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmsleu.vi v10, v8, 5
+; VLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v10, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmsleu.nxv4i32.i32(<vscale x 4 x i32> %a, i32 5, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
+
+define <vscale x 4 x i1> @vmsleu_vx(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, i32 %c, iXLen %vl) {
+; NOVLOPT-LABEL: vmsleu_vx:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmsleu.vx v10, v8, a0
+; NOVLOPT-NEXT: vsetvli zero, a1, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v10, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmsleu_vx:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmsleu.vx v10, v8, a0
+; VLOPT-NEXT: vsetvli zero, a1, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v10, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmsleu.nxv4i32.i32(<vscale x 4 x i32> %a, i32 %c, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
+
+define <vscale x 4 x i1> @vmsleu_vv(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, <vscale x 4 x i32> %c, iXLen %vl) {
+; NOVLOPT-LABEL: vmsleu_vv:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmsleu.vv v12, v8, v10
+; NOVLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v12, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmsleu_vv:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmsleu.vv v12, v8, v10
+; VLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v12, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmsleu.nxv4i32.nxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %c, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
+
+define <vscale x 4 x i1> @vmsle_vi(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, iXLen %vl) {
+; NOVLOPT-LABEL: vmsle_vi:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmsle.vi v10, v8, 5
+; NOVLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v10, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmsle_vi:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmsle.vi v10, v8, 5
+; VLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v10, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmsle.nxv4i32.i32(<vscale x 4 x i32> %a, i32 5, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
+
+define <vscale x 4 x i1> @vmsle_vx(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, i32 %c, iXLen %vl) {
+; NOVLOPT-LABEL: vmsle_vx:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmsle.vx v10, v8, a0
+; NOVLOPT-NEXT: vsetvli zero, a1, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v10, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmsle_vx:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmsle.vx v10, v8, a0
+; VLOPT-NEXT: vsetvli zero, a1, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v10, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmsle.nxv4i32.i32(<vscale x 4 x i32> %a, i32 %c, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
+
+define <vscale x 4 x i1> @vmsle_vv(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, <vscale x 4 x i32> %c, iXLen %vl) {
+; NOVLOPT-LABEL: vmsle_vv:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmsle.vv v12, v8, v10
+; NOVLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v12, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmsle_vv:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmsle.vv v12, v8, v10
+; VLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v12, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmsle.nxv4i32.nxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %c, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
+
+define <vscale x 4 x i1> @vmsgtu_vi(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, iXLen %vl) {
+; NOVLOPT-LABEL: vmsgtu_vi:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmsgtu.vi v10, v8, 5
+; NOVLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v10, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmsgtu_vi:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmsgtu.vi v10, v8, 5
+; VLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v10, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmsgtu.nxv4i32.i32(<vscale x 4 x i32> %a, i32 5, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
+
+define <vscale x 4 x i1> @vmsgtu_vx(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, i32 %c, iXLen %vl) {
+; NOVLOPT-LABEL: vmsgtu_vx:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmsgtu.vx v10, v8, a0
+; NOVLOPT-NEXT: vsetvli zero, a1, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v10, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmsgtu_vx:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmsgtu.vx v10, v8, a0
+; VLOPT-NEXT: vsetvli zero, a1, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v10, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmsgtu.nxv4i32.i32(<vscale x 4 x i32> %a, i32 %c, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
+
+define <vscale x 4 x i1> @vmsgt_vi(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, iXLen %vl) {
+; NOVLOPT-LABEL: vmsgt_vi:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmsgt.vi v10, v8, 5
+; NOVLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v10, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmsgt_vi:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a1, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmsgt.vi v10, v8, 5
+; VLOPT-NEXT: vsetvli zero, a0, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v10, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmsgt.nxv4i32.i32(<vscale x 4 x i32> %a, i32 5, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
+
+define <vscale x 4 x i1> @vmsgt_vx(<vscale x 4 x i32> %a, <vscale x 4 x i1> %b, i32 %c, iXLen %vl) {
+; NOVLOPT-LABEL: vmsgt_vx:
+; NOVLOPT: # %bb.0:
+; NOVLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
+; NOVLOPT-NEXT: vmsgt.vx v10, v8, a0
+; NOVLOPT-NEXT: vsetvli zero, a1, e8, mf2, ta, ma
+; NOVLOPT-NEXT: vmand.mm v0, v10, v0
+; NOVLOPT-NEXT: ret
+;
+; VLOPT-LABEL: vmsgt_vx:
+; VLOPT: # %bb.0:
+; VLOPT-NEXT: vsetvli a2, zero, e32, m2, ta, ma
+; VLOPT-NEXT: vmsgt.vx v10, v8, a0
+; VLOPT-NEXT: vsetvli zero, a1, e8, mf2, ta, ma
+; VLOPT-NEXT: vmand.mm v0, v10, v0
+; VLOPT-NEXT: ret
+ %1 = call <vscale x 4 x i1> @llvm.riscv.vmsgt.nxv4i32.i32(<vscale x 4 x i32> %a, i32 %c, iXLen -1)
+ %2 = call <vscale x 4 x i1> @llvm.riscv.vmand.nxv4i1(<vscale x 4 x i1> %1, <vscale x 4 x i1> %b, iXLen %vl)
+ ret <vscale x 4 x i1> %2
+}
define <vscale x 4 x i32> @vminu_vv(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl) {
; NOVLOPT-LABEL: vminu_vv:
diff --git a/llvm/test/CodeGen/RISCV/rvv/vl-opt-op-info.mir b/llvm/test/CodeGen/RISCV/rvv/vl-opt-op-info.mir
index f1e7bb446482e1..165f1ebfa5aa7e 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vl-opt-op-info.mir
+++ b/llvm/test/CodeGen/RISCV/rvv/vl-opt-op-info.mir
@@ -602,4 +602,83 @@ body: |
%x:vmv0 = PseudoVMAND_MM_M1 $noreg, $noreg, -1, 0
%y:vrnov0 = PseudoVADD_VV_MF2_MASK $noreg, $noreg, $noreg, %x, 1, 3 /* e8 */, 0
...
-
+---
+name: vmcmp_vv
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: vmcmp_vv
+ ; CHECK: %x:vr = PseudoVMSEQ_VV_M1 $noreg, $noreg, 1, 0 /* e8 */
+ ; CHECK-NEXT: %y:vr = PseudoVMAND_MM_M1 $noreg, %x, 1, 0 /* e8 */
+ %x:vr = PseudoVMSEQ_VV_M1 $noreg, $noreg, -1, 0
+ %y:vr = PseudoVMAND_MM_M1 $noreg, %x, 1, 0
+...
+---
+name: vmcmp_vv_maskuser
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: vmcmp_vv_maskuser
+ ; CHECK: %x:vmv0 = PseudoVMSEQ_VV_M1 $noreg, $noreg, 1, 0 /* e8 */
+ ; CHECK-NEXT: %y:vrnov0 = PseudoVADD_VV_M1_MASK $noreg, $noreg, $noreg, %x, 1, 3 /* e8 */, 0 /* tu, mu */
+ %x:vmv0 = PseudoVMSEQ_VV_M1 $noreg, $noreg, -1, 0
+ %y:vrnov0 = PseudoVADD_VV_M1_MASK $noreg, $noreg, $noreg, %x, 1, 3 /* e8 */, 0
+...
+---
+name: vmcmp_vv_maskuser_incompatible_eew
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: vmcmp_vv_maskuser_incompatible_eew
+ ; CHECK: %x:vmv0 = PseudoVMSEQ_VV_M1 $noreg, $noreg, -1, 0 /* e8 */
+ ; CHECK-NEXT: %y:vrnov0 = PseudoVADD_VV_M1_MASK $noreg, $noreg, $noreg, %x, 1, 4 /* e16 */, 0 /* tu, mu */
+ %x:vmv0 = PseudoVMSEQ_VV_M1 $noreg, $noreg, -1, 0
+ %y:vrnov0 = PseudoVADD_VV_M1_MASK $noreg, $noreg, $noreg, %x, 1, 4 /* e16 */, 0
+...
+---
+name: vmcmp_vv_incompatible_emul
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: vmcmp_vv_incompatible_emul
+ ; CHECK: %x:vr = PseudoVMSEQ_VV_M1 $noreg, $noreg, -1, 0 /* e8 */
+ ; CHECK-NEXT: %y:vr = PseudoVMAND_MM_MF2 $noreg, %x, 1, 0 /* e8 */
+ %x:vr = PseudoVMSEQ_VV_M1 $noreg, $noreg, -1, 0
+ %y:vr = PseudoVMAND_MM_MF2 $noreg, %x, 1, 0
+...
+---
+name: vmcmp_vv_maskuser_incompaible_emul
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: vmcmp_vv_maskuser_incompaible_emul
+ ; CHECK: %x:vmv0 = PseudoVMSEQ_VV_M1 $noreg, $noreg, -1, 0 /* e8 */
+ ; CHECK-NEXT: %y:vrnov0 = PseudoVADD_VV_MF2_MASK $noreg, $noreg, $noreg, %x, 1, 3 /* e8 */, 0 /* tu, mu */
+ %x:vmv0 = PseudoVMSEQ_VV_M1 $noreg, $noreg, -1, 0
+ %y:vrnov0 = PseudoVADD_VV_MF2_MASK $noreg, $noreg, $noreg, %x, 1, 3 /* e8 */, 0
+...
+---
+name: vmcmp_vv_maskuser_larger_emul
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: vmcmp_vv_maskuser_larger_emul
+ ; CHECK: %x:vmv0 = PseudoVMSEQ_VV_M1 $noreg, $noreg, -1, 0 /* e8 */
+ ; CHECK-NEXT: %y:vrm2nov0 = PseudoVADD_VV_M2_MASK $noreg, $noreg, $noreg, %x, 1, 3 /* e8 */, 0 /* tu, mu */
+ %x:vmv0 = PseudoVMSEQ_VV_M1 $noreg, $noreg, -1, 0
+ %y:vrm2nov0 = PseudoVADD_VV_M2_MASK $noreg, $noreg, $noreg, %x, 1, 3 /* e8 */, 0
+...
+---
+name: vmcmp_vv_consumer_incompatible_eew
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: vmcmp_vv_consumer
+ ; CHECK: %x:vr = PseudoVADD_VV_M1 $noreg, $noreg, $noreg, -1, 3 /* e8 */, 0 /* tu, mu */
+ ; CHECK-NEXT: %y:vr = PseudoVMSEQ_VV_M1 $noreg, %x, 1, 0 /* e8 */
+ %x:vr = PseudoVADD_VV_M1 $noreg, $noreg, $noreg, -1, 3 /* e8 */, 0
+ %y:vr = PseudoVMSEQ_VV_M1 $noreg, %x, 1, 0
+...
+---
+name: vmcmp_vv_consumer_incompatible_emul
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: vmcmp_vv_consumer_incompatible_emul
+ ; CHECK: %x:vr = PseudoVADD_VV_M1 $noreg, $noreg, $noreg, -1, 3 /* e8 */, 0 /* tu, mu */
+ ; CHECK-NEXT: %y:vr = PseudoVMSEQ_VV_MF2 $noreg, %x, 1, 0 /* e8 */
+ %x:vr = PseudoVADD_VV_M1 $noreg, $noreg, $noreg, -1, 3 /* e8 */, 0
+ %y:vr = PseudoVMSEQ_VV_MF2 $noreg, %x, 1, 0
+...
More information about the llvm-commits
mailing list