[RISCV][VLOPT] Add support for Vector Fixed-Point Arithmetic Instructions

llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp

@@ -976,6 +976,17 @@ static bool isSupportedInstr(const MachineInstr &MI) {
   case RISCV::VMV_V_I:
   case RISCV::VMV_V_X:
   case RISCV::VMV_V_V:
+  // Vector Single-Width Saturating Add and Subtract
+  case RISCV::VSADDU_VV:
+  case RISCV::VSADDU_VX:
+  case RISCV::VSADDU_VI:
+  case RISCV::VSADD_VV:
+  case RISCV::VSADD_VX:
+  case RISCV::VSADD_VI:
+  case RISCV::VSSUBU_VV:
+  case RISCV::VSSUBU_VX:
+  case RISCV::VSSUB_VV:
+  case RISCV::VSSUB_VX:
   // Vector Single-Width Averaging Add and Subtract
   case RISCV::VAADDU_VV:
   case RISCV::VAADDU_VX:
@@ -985,6 +996,23 @@ static bool isSupportedInstr(const MachineInstr &MI) {
   case RISCV::VASUBU_VX:
   case RISCV::VASUB_VV:
   case RISCV::VASUB_VX:
+  // Vector Single-Width Fractional Multiply with Rounding and Saturation
+  case RISCV::VSMUL_VV:
+  case RISCV::VSMUL_VX:
+  // Vector Single-Width Scaling Shift Instructions
+  case RISCV::VSSRL_VV:
+  case RISCV::VSSRL_VX:
+  case RISCV::VSSRL_VI:
+  case RISCV::VSSRA_VV:
+  case RISCV::VSSRA_VX:
+  case RISCV::VSSRA_VI:
+  // Vector Narrowing Fixed-Point Clip Instructions
+  case RISCV::VNCLIPU_WV:
+  case RISCV::VNCLIPU_WX:
+  case RISCV::VNCLIPU_WI:
+  case RISCV::VNCLIP_WV:
+  case RISCV::VNCLIP_WX:
+  case RISCV::VNCLIP_WI:
 
   // Vector Crypto
   case RISCV::VWSLL_VI:
@@ -1173,8 +1201,16 @@ bool RISCVVLOptimizer::isCandidate(const MachineInstr &MI) const {
   const MCInstrDesc &Desc = MI.getDesc();
   if (!RISCVII::hasVLOp(Desc.TSFlags) || !RISCVII::hasSEWOp(Desc.TSFlags))
     return false;
-  if (MI.getNumDefs() != 1)
+
+  if (MI.getNumExplicitDefs() != 1)
+    return false;
+
+  // Some instructions have implicit defs e.g. $vxsat. If they might be read
+  // later then we can't reduce VL.
+  if (!MI.allImplicitDefsAreDead()) {
+    LLVM_DEBUG(dbgs() << "Not a candidate because has non-dead implicit def\n");
     return false;
+  }
 
   if (MI.mayRaiseFPException()) {
     LLVM_DEBUG(dbgs() << "Not a candidate because may raise FP exception\n");

llvm/test/CodeGen/RISCV/rvv/rvv-peephole-vmerge-vops.ll

@@ -894,9 +894,10 @@ define void @test_dag_loop() {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    vsetvli a0, zero, e8, m4, ta, ma
 ; CHECK-NEXT:    vmclr.m v0
+; CHECK-NEXT:    vsetivli zero, 0, e8, m4, ta, ma
 ; CHECK-NEXT:    vmv.v.i v8, 0
 ; CHECK-NEXT:    vmv.v.i v12, 0
-; CHECK-NEXT:    vsetivli zero, 0, e8, m4, tu, mu
+; CHECK-NEXT:    vsetvli zero, zero, e8, m4, tu, mu
 ; CHECK-NEXT:    vssubu.vx v12, v8, zero, v0.t
 ; CHECK-NEXT:    vsetvli zero, zero, e8, m4, ta, ma
 ; CHECK-NEXT:    vmseq.vv v0, v12, v8