[llvm] [RISCV] Improve cleanup phase of RISCV Insert VSETVLI pass (PR #67144)
Simeon K via llvm-commits
llvm-commits at lists.llvm.org
Mon Sep 25 02:57:43 PDT 2023
https://github.com/simeonkr updated https://github.com/llvm/llvm-project/pull/67144
>From 3272838f98137507e98c76269104159acc2d9f40 Mon Sep 17 00:00:00 2001
From: Simeon Krastnikov <simeon.krastnikov at imgtec.com>
Date: Tue, 19 Sep 2023 16:14:47 +0100
Subject: [PATCH 1/2] Improve cleanup phase of RISCV Insert VSETVLI pass
Make the pass better at eliminating redundant VSETVLIs in certain
scenarios, as demonstrated in the updated tests.
---
llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp | 111 ++++----
.../RISCV/rvv/fixed-vectors-masked-gather.ll | 252 ++++++------------
.../RISCV/rvv/vsetvli-insert-crossbb.ll | 3 +-
llvm/test/CodeGen/RISCV/rvv/vsetvli-insert.ll | 8 +-
.../CodeGen/RISCV/rvv/vsetvli-intrinsics.ll | 18 +-
5 files changed, 158 insertions(+), 234 deletions(-)
diff --git a/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp b/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp
index 5584fa8d503dbe4..138c28e4effabf6 100644
--- a/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp
+++ b/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp
@@ -1426,43 +1426,6 @@ static bool isNonZeroAVL(const MachineOperand &MO) {
return 0 != MO.getImm();
}
-// Return true if we can mutate PrevMI to match MI without changing any the
-// fields which would be observed.
-static bool canMutatePriorConfig(const MachineInstr &PrevMI,
- const MachineInstr &MI,
- const DemandedFields &Used) {
- // If the VL values aren't equal, return false if either a) the former is
- // demanded, or b) we can't rewrite the former to be the later for
- // implementation reasons.
- if (!isVLPreservingConfig(MI)) {
- if (Used.VLAny)
- return false;
-
- // We don't bother to handle the equally zero case here as it's largely
- // uninteresting.
- if (Used.VLZeroness) {
- if (isVLPreservingConfig(PrevMI))
- return false;
- if (!isNonZeroAVL(MI.getOperand(1)) ||
- !isNonZeroAVL(PrevMI.getOperand(1)))
- return false;
- }
-
- // TODO: Track whether the register is defined between
- // PrevMI and MI.
- if (MI.getOperand(1).isReg() &&
- RISCV::X0 != MI.getOperand(1).getReg())
- return false;
- }
-
- if (!PrevMI.getOperand(2).isImm() || !MI.getOperand(2).isImm())
- return false;
-
- auto PriorVType = PrevMI.getOperand(2).getImm();
- auto VType = MI.getOperand(2).getImm();
- return areCompatibleVTYPEs(PriorVType, VType, Used);
-}
-
void RISCVInsertVSETVLI::doLocalPostpass(MachineBasicBlock &MBB) {
MachineInstr *NextMI = nullptr;
// We can have arbitrary code in successors, so VL and VTYPE
@@ -1475,36 +1438,74 @@ void RISCVInsertVSETVLI::doLocalPostpass(MachineBasicBlock &MBB) {
if (!isVectorConfigInstr(MI)) {
doUnion(Used, getDemanded(MI, MRI));
+ // We can't handle the case when the source AVL
+ // register of *NextMI is defined after MI
+ if (NextMI && NextMI->getOperand(1).isReg() && MI.getOperand(0).isReg() &&
+ MI.getOperand(0).getReg() == NextMI->getOperand(1).getReg())
+ Used.demandVL();
continue;
}
Register VRegDef = MI.getOperand(0).getReg();
if (VRegDef != RISCV::X0 &&
- !(VRegDef.isVirtual() && MRI->use_nodbg_empty(VRegDef)))
- Used.demandVL();
+ !(VRegDef.isVirtual() && MRI->use_nodbg_empty(VRegDef))) {
+ for (MachineInstr &UserMI : MRI->use_nodbg_instructions(VRegDef)) {
+ // ignore uses by *NextMI since we will deal with them explicitly
+ if (!NextMI || !UserMI.isIdenticalTo(*NextMI)) {
+ Used.demandVL();
+ break;
+ }
+ }
+ }
if (NextMI) {
- if (!Used.usedVL() && !Used.usedVTYPE()) {
- ToDelete.push_back(&MI);
- // Leave NextMI unchanged
- continue;
- } else if (canMutatePriorConfig(MI, *NextMI, Used)) {
- if (!isVLPreservingConfig(*NextMI)) {
- MI.getOperand(0).setReg(NextMI->getOperand(0).getReg());
- MI.getOperand(0).setIsDead(false);
- if (NextMI->getOperand(1).isImm())
- MI.getOperand(1).ChangeToImmediate(NextMI->getOperand(1).getImm());
- else
- MI.getOperand(1).ChangeToRegister(NextMI->getOperand(1).getReg(), false);
- MI.setDesc(NextMI->getDesc());
+ auto NextVRegDef = NextMI->getOperand(0).getReg();
+ auto &AVLOp = MI.getOperand(1), &NextAVLOp = NextMI->getOperand(1);
+ auto &VLTypeOp = MI.getOperand(2), &NextVLTypeOp = NextMI->getOperand(2);
+ if (areCompatibleVTYPEs(VLTypeOp.getImm(), NextVLTypeOp.getImm(), Used)) {
+ bool VLUsed =
+ Used.VLAny || (Used.VLZeroness &&
+ (!isNonZeroAVL(AVLOp) || !isNonZeroAVL(NextAVLOp)));
+ if (isVLPreservingConfig(*NextMI) || !VLUsed) {
+ // set VLType of MI to that of *NextMI
+ VLTypeOp.setImm(NextVLTypeOp.getImm());
+ // remove *NextMI and potentially replace its uses with MI
+ ToDelete.push_back(NextMI);
+ if (NextVRegDef != RISCV::X0 && !MRI->use_nodbg_empty(NextVRegDef)) {
+ if (VRegDef != RISCV::X0)
+ MRI->replaceRegWith(NextVRegDef, VRegDef);
+ else {
+ MI.getOperand(0).ChangeToRegister(
+ MRI->createVirtualRegister(&RISCV::GPRRegClass), true);
+ MRI->replaceRegWith(NextVRegDef, MI.getOperand(0).getReg());
+ }
+ }
+ // is the AVL op of *NextMI a non-zero reg storing the result of MI?
+ bool KeepAVL = NextAVLOp.isReg() && NextAVLOp.getReg() == VRegDef &&
+ VRegDef != RISCV::X0;
+ // set AVL of MI to that of *NextMI whenever it makes sense to do so
+ if (!isVLPreservingConfig(*NextMI) && !KeepAVL) {
+ if (NextAVLOp.isImm())
+ AVLOp.ChangeToImmediate(NextAVLOp.getImm());
+ else {
+ AVLOp.ChangeToRegister(NextAVLOp.getReg(), false);
+ // ensure that *MI did not inadvertently become VL-preserving
+ if (VRegDef == RISCV::X0 && AVLOp.getReg() == RISCV::X0)
+ MI.getOperand(0).ChangeToRegister(
+ MRI->createVirtualRegister(&RISCV::GPRRegClass), true);
+ }
+ MI.setDesc(NextMI->getDesc());
+ }
}
- MI.getOperand(2).setImm(NextMI->getOperand(2).getImm());
- ToDelete.push_back(NextMI);
- // fallthrough
}
}
NextMI = &MI;
Used = getDemanded(MI, MRI);
+ // ignore such uses since we can handle them explicitly
+ if (VRegDef == RISCV::X0) {
+ Used.VLAny = false;
+ Used.VLZeroness = false;
+ }
}
for (auto *MI : ToDelete)
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
index 49724cbc4418252..1bb197c84799b83 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
@@ -2530,9 +2530,8 @@ define <8 x i32> @mgather_baseidx_v8i8_v8i32(ptr %base, <8 x i8> %idxs, <8 x i1>
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vsetivli zero, 2, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 1
; RV64ZVE32F-NEXT: .LBB35_4: # %else2
; RV64ZVE32F-NEXT: vsetivli zero, 4, e8, mf2, ta, ma
@@ -2557,9 +2556,8 @@ define <8 x i32> @mgather_baseidx_v8i8_v8i32(ptr %base, <8 x i8> %idxs, <8 x i1>
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v8, a2
; RV64ZVE32F-NEXT: vsetivli zero, 6, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v8, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 5
; RV64ZVE32F-NEXT: .LBB35_9: # %else14
; RV64ZVE32F-NEXT: vsetivli zero, 2, e8, mf4, ta, ma
@@ -2577,9 +2575,8 @@ define <8 x i32> @mgather_baseidx_v8i8_v8i32(ptr %base, <8 x i8> %idxs, <8 x i1>
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vsetivli zero, 3, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 2
; RV64ZVE32F-NEXT: andi a2, a1, 8
; RV64ZVE32F-NEXT: beqz a2, .LBB35_6
@@ -2590,9 +2587,8 @@ define <8 x i32> @mgather_baseidx_v8i8_v8i32(ptr %base, <8 x i8> %idxs, <8 x i1>
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vsetivli zero, 4, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 3
; RV64ZVE32F-NEXT: andi a2, a1, 16
; RV64ZVE32F-NEXT: beqz a2, .LBB35_7
@@ -2613,9 +2609,8 @@ define <8 x i32> @mgather_baseidx_v8i8_v8i32(ptr %base, <8 x i8> %idxs, <8 x i1>
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vsetivli zero, 7, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 6
; RV64ZVE32F-NEXT: andi a1, a1, -128
; RV64ZVE32F-NEXT: beqz a1, .LBB35_11
@@ -2626,9 +2621,8 @@ define <8 x i32> @mgather_baseidx_v8i8_v8i32(ptr %base, <8 x i8> %idxs, <8 x i1>
; RV64ZVE32F-NEXT: slli a1, a1, 2
; RV64ZVE32F-NEXT: add a0, a0, a1
; RV64ZVE32F-NEXT: lw a0, 0(a0)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v8, a0
; RV64ZVE32F-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV64ZVE32F-NEXT: vmv.s.x v8, a0
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 7
; RV64ZVE32F-NEXT: vmv2r.v v8, v10
; RV64ZVE32F-NEXT: ret
@@ -2680,9 +2674,8 @@ define <8 x i32> @mgather_baseidx_sext_v8i8_v8i32(ptr %base, <8 x i8> %idxs, <8
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vsetivli zero, 2, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 1
; RV64ZVE32F-NEXT: .LBB36_4: # %else2
; RV64ZVE32F-NEXT: vsetivli zero, 4, e8, mf2, ta, ma
@@ -2707,9 +2700,8 @@ define <8 x i32> @mgather_baseidx_sext_v8i8_v8i32(ptr %base, <8 x i8> %idxs, <8
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v8, a2
; RV64ZVE32F-NEXT: vsetivli zero, 6, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v8, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 5
; RV64ZVE32F-NEXT: .LBB36_9: # %else14
; RV64ZVE32F-NEXT: vsetivli zero, 2, e8, mf4, ta, ma
@@ -2727,9 +2719,8 @@ define <8 x i32> @mgather_baseidx_sext_v8i8_v8i32(ptr %base, <8 x i8> %idxs, <8
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vsetivli zero, 3, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 2
; RV64ZVE32F-NEXT: andi a2, a1, 8
; RV64ZVE32F-NEXT: beqz a2, .LBB36_6
@@ -2740,9 +2731,8 @@ define <8 x i32> @mgather_baseidx_sext_v8i8_v8i32(ptr %base, <8 x i8> %idxs, <8
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vsetivli zero, 4, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 3
; RV64ZVE32F-NEXT: andi a2, a1, 16
; RV64ZVE32F-NEXT: beqz a2, .LBB36_7
@@ -2763,9 +2753,8 @@ define <8 x i32> @mgather_baseidx_sext_v8i8_v8i32(ptr %base, <8 x i8> %idxs, <8
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vsetivli zero, 7, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 6
; RV64ZVE32F-NEXT: andi a1, a1, -128
; RV64ZVE32F-NEXT: beqz a1, .LBB36_11
@@ -2776,9 +2765,8 @@ define <8 x i32> @mgather_baseidx_sext_v8i8_v8i32(ptr %base, <8 x i8> %idxs, <8
; RV64ZVE32F-NEXT: slli a1, a1, 2
; RV64ZVE32F-NEXT: add a0, a0, a1
; RV64ZVE32F-NEXT: lw a0, 0(a0)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v8, a0
; RV64ZVE32F-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV64ZVE32F-NEXT: vmv.s.x v8, a0
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 7
; RV64ZVE32F-NEXT: vmv2r.v v8, v10
; RV64ZVE32F-NEXT: ret
@@ -2834,9 +2822,8 @@ define <8 x i32> @mgather_baseidx_zext_v8i8_v8i32(ptr %base, <8 x i8> %idxs, <8
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vsetivli zero, 2, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 1
; RV64ZVE32F-NEXT: .LBB37_4: # %else2
; RV64ZVE32F-NEXT: vsetivli zero, 4, e8, mf2, ta, ma
@@ -2862,9 +2849,8 @@ define <8 x i32> @mgather_baseidx_zext_v8i8_v8i32(ptr %base, <8 x i8> %idxs, <8
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v8, a2
; RV64ZVE32F-NEXT: vsetivli zero, 6, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v8, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 5
; RV64ZVE32F-NEXT: .LBB37_9: # %else14
; RV64ZVE32F-NEXT: vsetivli zero, 2, e8, mf4, ta, ma
@@ -2883,9 +2869,8 @@ define <8 x i32> @mgather_baseidx_zext_v8i8_v8i32(ptr %base, <8 x i8> %idxs, <8
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vsetivli zero, 3, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 2
; RV64ZVE32F-NEXT: andi a2, a1, 8
; RV64ZVE32F-NEXT: beqz a2, .LBB37_6
@@ -2897,9 +2882,8 @@ define <8 x i32> @mgather_baseidx_zext_v8i8_v8i32(ptr %base, <8 x i8> %idxs, <8
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v8, a2
; RV64ZVE32F-NEXT: vsetivli zero, 4, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v8, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 3
; RV64ZVE32F-NEXT: andi a2, a1, 16
; RV64ZVE32F-NEXT: beqz a2, .LBB37_7
@@ -2922,9 +2906,8 @@ define <8 x i32> @mgather_baseidx_zext_v8i8_v8i32(ptr %base, <8 x i8> %idxs, <8
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vsetivli zero, 7, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 6
; RV64ZVE32F-NEXT: andi a1, a1, -128
; RV64ZVE32F-NEXT: beqz a1, .LBB37_11
@@ -2936,9 +2919,8 @@ define <8 x i32> @mgather_baseidx_zext_v8i8_v8i32(ptr %base, <8 x i8> %idxs, <8
; RV64ZVE32F-NEXT: slli a1, a1, 2
; RV64ZVE32F-NEXT: add a0, a0, a1
; RV64ZVE32F-NEXT: lw a0, 0(a0)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v8, a0
; RV64ZVE32F-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV64ZVE32F-NEXT: vmv.s.x v8, a0
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 7
; RV64ZVE32F-NEXT: vmv2r.v v8, v10
; RV64ZVE32F-NEXT: ret
@@ -2992,9 +2974,8 @@ define <8 x i32> @mgather_baseidx_v8i16_v8i32(ptr %base, <8 x i16> %idxs, <8 x i
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vsetivli zero, 2, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 1
; RV64ZVE32F-NEXT: .LBB38_4: # %else2
; RV64ZVE32F-NEXT: vsetivli zero, 4, e16, m1, ta, ma
@@ -3019,9 +3000,8 @@ define <8 x i32> @mgather_baseidx_v8i16_v8i32(ptr %base, <8 x i16> %idxs, <8 x i
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v8, a2
; RV64ZVE32F-NEXT: vsetivli zero, 6, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v8, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 5
; RV64ZVE32F-NEXT: .LBB38_9: # %else14
; RV64ZVE32F-NEXT: vsetivli zero, 2, e16, mf2, ta, ma
@@ -3039,9 +3019,8 @@ define <8 x i32> @mgather_baseidx_v8i16_v8i32(ptr %base, <8 x i16> %idxs, <8 x i
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vsetivli zero, 3, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 2
; RV64ZVE32F-NEXT: andi a2, a1, 8
; RV64ZVE32F-NEXT: beqz a2, .LBB38_6
@@ -3052,9 +3031,8 @@ define <8 x i32> @mgather_baseidx_v8i16_v8i32(ptr %base, <8 x i16> %idxs, <8 x i
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vsetivli zero, 4, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 3
; RV64ZVE32F-NEXT: andi a2, a1, 16
; RV64ZVE32F-NEXT: beqz a2, .LBB38_7
@@ -3075,9 +3053,8 @@ define <8 x i32> @mgather_baseidx_v8i16_v8i32(ptr %base, <8 x i16> %idxs, <8 x i
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vsetivli zero, 7, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 6
; RV64ZVE32F-NEXT: andi a1, a1, -128
; RV64ZVE32F-NEXT: beqz a1, .LBB38_11
@@ -3088,9 +3065,8 @@ define <8 x i32> @mgather_baseidx_v8i16_v8i32(ptr %base, <8 x i16> %idxs, <8 x i
; RV64ZVE32F-NEXT: slli a1, a1, 2
; RV64ZVE32F-NEXT: add a0, a0, a1
; RV64ZVE32F-NEXT: lw a0, 0(a0)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v8, a0
; RV64ZVE32F-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV64ZVE32F-NEXT: vmv.s.x v8, a0
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 7
; RV64ZVE32F-NEXT: vmv2r.v v8, v10
; RV64ZVE32F-NEXT: ret
@@ -3143,9 +3119,8 @@ define <8 x i32> @mgather_baseidx_sext_v8i16_v8i32(ptr %base, <8 x i16> %idxs, <
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vsetivli zero, 2, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 1
; RV64ZVE32F-NEXT: .LBB39_4: # %else2
; RV64ZVE32F-NEXT: vsetivli zero, 4, e16, m1, ta, ma
@@ -3170,9 +3145,8 @@ define <8 x i32> @mgather_baseidx_sext_v8i16_v8i32(ptr %base, <8 x i16> %idxs, <
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v8, a2
; RV64ZVE32F-NEXT: vsetivli zero, 6, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v8, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 5
; RV64ZVE32F-NEXT: .LBB39_9: # %else14
; RV64ZVE32F-NEXT: vsetivli zero, 2, e16, mf2, ta, ma
@@ -3190,9 +3164,8 @@ define <8 x i32> @mgather_baseidx_sext_v8i16_v8i32(ptr %base, <8 x i16> %idxs, <
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vsetivli zero, 3, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 2
; RV64ZVE32F-NEXT: andi a2, a1, 8
; RV64ZVE32F-NEXT: beqz a2, .LBB39_6
@@ -3203,9 +3176,8 @@ define <8 x i32> @mgather_baseidx_sext_v8i16_v8i32(ptr %base, <8 x i16> %idxs, <
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vsetivli zero, 4, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 3
; RV64ZVE32F-NEXT: andi a2, a1, 16
; RV64ZVE32F-NEXT: beqz a2, .LBB39_7
@@ -3226,9 +3198,8 @@ define <8 x i32> @mgather_baseidx_sext_v8i16_v8i32(ptr %base, <8 x i16> %idxs, <
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vsetivli zero, 7, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a2
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 6
; RV64ZVE32F-NEXT: andi a1, a1, -128
; RV64ZVE32F-NEXT: beqz a1, .LBB39_11
@@ -3239,9 +3210,8 @@ define <8 x i32> @mgather_baseidx_sext_v8i16_v8i32(ptr %base, <8 x i16> %idxs, <
; RV64ZVE32F-NEXT: slli a1, a1, 2
; RV64ZVE32F-NEXT: add a0, a0, a1
; RV64ZVE32F-NEXT: lw a0, 0(a0)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v8, a0
; RV64ZVE32F-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV64ZVE32F-NEXT: vmv.s.x v8, a0
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 7
; RV64ZVE32F-NEXT: vmv2r.v v8, v10
; RV64ZVE32F-NEXT: ret
@@ -3298,9 +3268,8 @@ define <8 x i32> @mgather_baseidx_zext_v8i16_v8i32(ptr %base, <8 x i16> %idxs, <
; RV64ZVE32F-NEXT: slli a3, a3, 2
; RV64ZVE32F-NEXT: add a3, a0, a3
; RV64ZVE32F-NEXT: lw a3, 0(a3)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a3
; RV64ZVE32F-NEXT: vsetivli zero, 2, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a3
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 1
; RV64ZVE32F-NEXT: .LBB40_4: # %else2
; RV64ZVE32F-NEXT: vsetivli zero, 4, e16, m1, ta, ma
@@ -3326,9 +3295,8 @@ define <8 x i32> @mgather_baseidx_zext_v8i16_v8i32(ptr %base, <8 x i16> %idxs, <
; RV64ZVE32F-NEXT: slli a3, a3, 2
; RV64ZVE32F-NEXT: add a3, a0, a3
; RV64ZVE32F-NEXT: lw a3, 0(a3)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v8, a3
; RV64ZVE32F-NEXT: vsetivli zero, 6, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v8, a3
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 5
; RV64ZVE32F-NEXT: .LBB40_9: # %else14
; RV64ZVE32F-NEXT: vsetivli zero, 2, e16, mf2, ta, ma
@@ -3347,9 +3315,8 @@ define <8 x i32> @mgather_baseidx_zext_v8i16_v8i32(ptr %base, <8 x i16> %idxs, <
; RV64ZVE32F-NEXT: slli a3, a3, 2
; RV64ZVE32F-NEXT: add a3, a0, a3
; RV64ZVE32F-NEXT: lw a3, 0(a3)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a3
; RV64ZVE32F-NEXT: vsetivli zero, 3, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a3
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 2
; RV64ZVE32F-NEXT: andi a3, a2, 8
; RV64ZVE32F-NEXT: beqz a3, .LBB40_6
@@ -3361,9 +3328,8 @@ define <8 x i32> @mgather_baseidx_zext_v8i16_v8i32(ptr %base, <8 x i16> %idxs, <
; RV64ZVE32F-NEXT: slli a3, a3, 2
; RV64ZVE32F-NEXT: add a3, a0, a3
; RV64ZVE32F-NEXT: lw a3, 0(a3)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a3
; RV64ZVE32F-NEXT: vsetivli zero, 4, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a3
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 3
; RV64ZVE32F-NEXT: andi a3, a2, 16
; RV64ZVE32F-NEXT: beqz a3, .LBB40_7
@@ -3386,9 +3352,8 @@ define <8 x i32> @mgather_baseidx_zext_v8i16_v8i32(ptr %base, <8 x i16> %idxs, <
; RV64ZVE32F-NEXT: slli a3, a3, 2
; RV64ZVE32F-NEXT: add a3, a0, a3
; RV64ZVE32F-NEXT: lw a3, 0(a3)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v12, a3
; RV64ZVE32F-NEXT: vsetivli zero, 7, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v12, a3
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 6
; RV64ZVE32F-NEXT: andi a2, a2, -128
; RV64ZVE32F-NEXT: beqz a2, .LBB40_11
@@ -3400,9 +3365,8 @@ define <8 x i32> @mgather_baseidx_zext_v8i16_v8i32(ptr %base, <8 x i16> %idxs, <
; RV64ZVE32F-NEXT: slli a1, a1, 2
; RV64ZVE32F-NEXT: add a0, a0, a1
; RV64ZVE32F-NEXT: lw a0, 0(a0)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v8, a0
; RV64ZVE32F-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV64ZVE32F-NEXT: vmv.s.x v8, a0
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 7
; RV64ZVE32F-NEXT: vmv2r.v v8, v10
; RV64ZVE32F-NEXT: ret
@@ -8333,9 +8297,8 @@ define <8 x float> @mgather_baseidx_v8i8_v8f32(ptr %base, <8 x i8> %idxs, <8 x i
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 2, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 1
; RV64ZVE32F-NEXT: .LBB74_4: # %else2
; RV64ZVE32F-NEXT: vsetivli zero, 4, e8, mf2, ta, ma
@@ -8360,9 +8323,8 @@ define <8 x float> @mgather_baseidx_v8i8_v8f32(ptr %base, <8 x i8> %idxs, <8 x i
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 6, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 5
; RV64ZVE32F-NEXT: .LBB74_9: # %else14
; RV64ZVE32F-NEXT: vsetivli zero, 2, e8, mf4, ta, ma
@@ -8380,9 +8342,8 @@ define <8 x float> @mgather_baseidx_v8i8_v8f32(ptr %base, <8 x i8> %idxs, <8 x i
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 3, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 2
; RV64ZVE32F-NEXT: andi a2, a1, 8
; RV64ZVE32F-NEXT: beqz a2, .LBB74_6
@@ -8393,9 +8354,8 @@ define <8 x float> @mgather_baseidx_v8i8_v8f32(ptr %base, <8 x i8> %idxs, <8 x i
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 4, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 3
; RV64ZVE32F-NEXT: andi a2, a1, 16
; RV64ZVE32F-NEXT: beqz a2, .LBB74_7
@@ -8416,9 +8376,8 @@ define <8 x float> @mgather_baseidx_v8i8_v8f32(ptr %base, <8 x i8> %idxs, <8 x i
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 7, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 6
; RV64ZVE32F-NEXT: andi a1, a1, -128
; RV64ZVE32F-NEXT: beqz a1, .LBB74_11
@@ -8429,9 +8388,8 @@ define <8 x float> @mgather_baseidx_v8i8_v8f32(ptr %base, <8 x i8> %idxs, <8 x i
; RV64ZVE32F-NEXT: slli a1, a1, 2
; RV64ZVE32F-NEXT: add a0, a0, a1
; RV64ZVE32F-NEXT: flw fa5, 0(a0)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 7
; RV64ZVE32F-NEXT: vmv2r.v v8, v10
; RV64ZVE32F-NEXT: ret
@@ -8483,9 +8441,8 @@ define <8 x float> @mgather_baseidx_sext_v8i8_v8f32(ptr %base, <8 x i8> %idxs, <
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 2, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 1
; RV64ZVE32F-NEXT: .LBB75_4: # %else2
; RV64ZVE32F-NEXT: vsetivli zero, 4, e8, mf2, ta, ma
@@ -8510,9 +8467,8 @@ define <8 x float> @mgather_baseidx_sext_v8i8_v8f32(ptr %base, <8 x i8> %idxs, <
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 6, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 5
; RV64ZVE32F-NEXT: .LBB75_9: # %else14
; RV64ZVE32F-NEXT: vsetivli zero, 2, e8, mf4, ta, ma
@@ -8530,9 +8486,8 @@ define <8 x float> @mgather_baseidx_sext_v8i8_v8f32(ptr %base, <8 x i8> %idxs, <
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 3, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 2
; RV64ZVE32F-NEXT: andi a2, a1, 8
; RV64ZVE32F-NEXT: beqz a2, .LBB75_6
@@ -8543,9 +8498,8 @@ define <8 x float> @mgather_baseidx_sext_v8i8_v8f32(ptr %base, <8 x i8> %idxs, <
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 4, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 3
; RV64ZVE32F-NEXT: andi a2, a1, 16
; RV64ZVE32F-NEXT: beqz a2, .LBB75_7
@@ -8566,9 +8520,8 @@ define <8 x float> @mgather_baseidx_sext_v8i8_v8f32(ptr %base, <8 x i8> %idxs, <
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 7, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 6
; RV64ZVE32F-NEXT: andi a1, a1, -128
; RV64ZVE32F-NEXT: beqz a1, .LBB75_11
@@ -8579,9 +8532,8 @@ define <8 x float> @mgather_baseidx_sext_v8i8_v8f32(ptr %base, <8 x i8> %idxs, <
; RV64ZVE32F-NEXT: slli a1, a1, 2
; RV64ZVE32F-NEXT: add a0, a0, a1
; RV64ZVE32F-NEXT: flw fa5, 0(a0)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 7
; RV64ZVE32F-NEXT: vmv2r.v v8, v10
; RV64ZVE32F-NEXT: ret
@@ -8637,9 +8589,8 @@ define <8 x float> @mgather_baseidx_zext_v8i8_v8f32(ptr %base, <8 x i8> %idxs, <
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 2, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 1
; RV64ZVE32F-NEXT: .LBB76_4: # %else2
; RV64ZVE32F-NEXT: vsetivli zero, 4, e8, mf2, ta, ma
@@ -8665,9 +8616,8 @@ define <8 x float> @mgather_baseidx_zext_v8i8_v8f32(ptr %base, <8 x i8> %idxs, <
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 6, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 5
; RV64ZVE32F-NEXT: .LBB76_9: # %else14
; RV64ZVE32F-NEXT: vsetivli zero, 2, e8, mf4, ta, ma
@@ -8686,9 +8636,8 @@ define <8 x float> @mgather_baseidx_zext_v8i8_v8f32(ptr %base, <8 x i8> %idxs, <
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 3, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 2
; RV64ZVE32F-NEXT: andi a2, a1, 8
; RV64ZVE32F-NEXT: beqz a2, .LBB76_6
@@ -8700,9 +8649,8 @@ define <8 x float> @mgather_baseidx_zext_v8i8_v8f32(ptr %base, <8 x i8> %idxs, <
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 4, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 3
; RV64ZVE32F-NEXT: andi a2, a1, 16
; RV64ZVE32F-NEXT: beqz a2, .LBB76_7
@@ -8725,9 +8673,8 @@ define <8 x float> @mgather_baseidx_zext_v8i8_v8f32(ptr %base, <8 x i8> %idxs, <
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 7, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 6
; RV64ZVE32F-NEXT: andi a1, a1, -128
; RV64ZVE32F-NEXT: beqz a1, .LBB76_11
@@ -8739,9 +8686,8 @@ define <8 x float> @mgather_baseidx_zext_v8i8_v8f32(ptr %base, <8 x i8> %idxs, <
; RV64ZVE32F-NEXT: slli a1, a1, 2
; RV64ZVE32F-NEXT: add a0, a0, a1
; RV64ZVE32F-NEXT: flw fa5, 0(a0)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 7
; RV64ZVE32F-NEXT: vmv2r.v v8, v10
; RV64ZVE32F-NEXT: ret
@@ -8795,9 +8741,8 @@ define <8 x float> @mgather_baseidx_v8i16_v8f32(ptr %base, <8 x i16> %idxs, <8 x
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 2, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 1
; RV64ZVE32F-NEXT: .LBB77_4: # %else2
; RV64ZVE32F-NEXT: vsetivli zero, 4, e16, m1, ta, ma
@@ -8822,9 +8767,8 @@ define <8 x float> @mgather_baseidx_v8i16_v8f32(ptr %base, <8 x i16> %idxs, <8 x
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 6, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 5
; RV64ZVE32F-NEXT: .LBB77_9: # %else14
; RV64ZVE32F-NEXT: vsetivli zero, 2, e16, mf2, ta, ma
@@ -8842,9 +8786,8 @@ define <8 x float> @mgather_baseidx_v8i16_v8f32(ptr %base, <8 x i16> %idxs, <8 x
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 3, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 2
; RV64ZVE32F-NEXT: andi a2, a1, 8
; RV64ZVE32F-NEXT: beqz a2, .LBB77_6
@@ -8855,9 +8798,8 @@ define <8 x float> @mgather_baseidx_v8i16_v8f32(ptr %base, <8 x i16> %idxs, <8 x
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 4, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 3
; RV64ZVE32F-NEXT: andi a2, a1, 16
; RV64ZVE32F-NEXT: beqz a2, .LBB77_7
@@ -8878,9 +8820,8 @@ define <8 x float> @mgather_baseidx_v8i16_v8f32(ptr %base, <8 x i16> %idxs, <8 x
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 7, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 6
; RV64ZVE32F-NEXT: andi a1, a1, -128
; RV64ZVE32F-NEXT: beqz a1, .LBB77_11
@@ -8891,9 +8832,8 @@ define <8 x float> @mgather_baseidx_v8i16_v8f32(ptr %base, <8 x i16> %idxs, <8 x
; RV64ZVE32F-NEXT: slli a1, a1, 2
; RV64ZVE32F-NEXT: add a0, a0, a1
; RV64ZVE32F-NEXT: flw fa5, 0(a0)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 7
; RV64ZVE32F-NEXT: vmv2r.v v8, v10
; RV64ZVE32F-NEXT: ret
@@ -8946,9 +8886,8 @@ define <8 x float> @mgather_baseidx_sext_v8i16_v8f32(ptr %base, <8 x i16> %idxs,
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 2, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 1
; RV64ZVE32F-NEXT: .LBB78_4: # %else2
; RV64ZVE32F-NEXT: vsetivli zero, 4, e16, m1, ta, ma
@@ -8973,9 +8912,8 @@ define <8 x float> @mgather_baseidx_sext_v8i16_v8f32(ptr %base, <8 x i16> %idxs,
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 6, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 5
; RV64ZVE32F-NEXT: .LBB78_9: # %else14
; RV64ZVE32F-NEXT: vsetivli zero, 2, e16, mf2, ta, ma
@@ -8993,9 +8931,8 @@ define <8 x float> @mgather_baseidx_sext_v8i16_v8f32(ptr %base, <8 x i16> %idxs,
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 3, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 2
; RV64ZVE32F-NEXT: andi a2, a1, 8
; RV64ZVE32F-NEXT: beqz a2, .LBB78_6
@@ -9006,9 +8943,8 @@ define <8 x float> @mgather_baseidx_sext_v8i16_v8f32(ptr %base, <8 x i16> %idxs,
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 4, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 3
; RV64ZVE32F-NEXT: andi a2, a1, 16
; RV64ZVE32F-NEXT: beqz a2, .LBB78_7
@@ -9029,9 +8965,8 @@ define <8 x float> @mgather_baseidx_sext_v8i16_v8f32(ptr %base, <8 x i16> %idxs,
; RV64ZVE32F-NEXT: slli a2, a2, 2
; RV64ZVE32F-NEXT: add a2, a0, a2
; RV64ZVE32F-NEXT: flw fa5, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 7, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 6
; RV64ZVE32F-NEXT: andi a1, a1, -128
; RV64ZVE32F-NEXT: beqz a1, .LBB78_11
@@ -9042,9 +8977,8 @@ define <8 x float> @mgather_baseidx_sext_v8i16_v8f32(ptr %base, <8 x i16> %idxs,
; RV64ZVE32F-NEXT: slli a1, a1, 2
; RV64ZVE32F-NEXT: add a0, a0, a1
; RV64ZVE32F-NEXT: flw fa5, 0(a0)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 7
; RV64ZVE32F-NEXT: vmv2r.v v8, v10
; RV64ZVE32F-NEXT: ret
@@ -9101,9 +9035,8 @@ define <8 x float> @mgather_baseidx_zext_v8i16_v8f32(ptr %base, <8 x i16> %idxs,
; RV64ZVE32F-NEXT: slli a3, a3, 2
; RV64ZVE32F-NEXT: add a3, a0, a3
; RV64ZVE32F-NEXT: flw fa5, 0(a3)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 2, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 1
; RV64ZVE32F-NEXT: .LBB79_4: # %else2
; RV64ZVE32F-NEXT: vsetivli zero, 4, e16, m1, ta, ma
@@ -9129,9 +9062,8 @@ define <8 x float> @mgather_baseidx_zext_v8i16_v8f32(ptr %base, <8 x i16> %idxs,
; RV64ZVE32F-NEXT: slli a3, a3, 2
; RV64ZVE32F-NEXT: add a3, a0, a3
; RV64ZVE32F-NEXT: flw fa5, 0(a3)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 6, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 5
; RV64ZVE32F-NEXT: .LBB79_9: # %else14
; RV64ZVE32F-NEXT: vsetivli zero, 2, e16, mf2, ta, ma
@@ -9150,9 +9082,8 @@ define <8 x float> @mgather_baseidx_zext_v8i16_v8f32(ptr %base, <8 x i16> %idxs,
; RV64ZVE32F-NEXT: slli a3, a3, 2
; RV64ZVE32F-NEXT: add a3, a0, a3
; RV64ZVE32F-NEXT: flw fa5, 0(a3)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 3, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 2
; RV64ZVE32F-NEXT: andi a3, a2, 8
; RV64ZVE32F-NEXT: beqz a3, .LBB79_6
@@ -9164,9 +9095,8 @@ define <8 x float> @mgather_baseidx_zext_v8i16_v8f32(ptr %base, <8 x i16> %idxs,
; RV64ZVE32F-NEXT: slli a3, a3, 2
; RV64ZVE32F-NEXT: add a3, a0, a3
; RV64ZVE32F-NEXT: flw fa5, 0(a3)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 4, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 3
; RV64ZVE32F-NEXT: andi a3, a2, 16
; RV64ZVE32F-NEXT: beqz a3, .LBB79_7
@@ -9189,9 +9119,8 @@ define <8 x float> @mgather_baseidx_zext_v8i16_v8f32(ptr %base, <8 x i16> %idxs,
; RV64ZVE32F-NEXT: slli a3, a3, 2
; RV64ZVE32F-NEXT: add a3, a0, a3
; RV64ZVE32F-NEXT: flw fa5, 0(a3)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 7, e32, m2, tu, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v12, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v12, 6
; RV64ZVE32F-NEXT: andi a2, a2, -128
; RV64ZVE32F-NEXT: beqz a2, .LBB79_11
@@ -9203,9 +9132,8 @@ define <8 x float> @mgather_baseidx_zext_v8i16_v8f32(ptr %base, <8 x i16> %idxs,
; RV64ZVE32F-NEXT: slli a1, a1, 2
; RV64ZVE32F-NEXT: add a0, a0, a1
; RV64ZVE32F-NEXT: flw fa5, 0(a0)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e32, m1, ta, ma
-; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; RV64ZVE32F-NEXT: vfmv.s.f v8, fa5
; RV64ZVE32F-NEXT: vslideup.vi v10, v8, 7
; RV64ZVE32F-NEXT: vmv2r.v v8, v10
; RV64ZVE32F-NEXT: ret
@@ -13485,9 +13413,8 @@ define <8 x i16> @mgather_strided_unaligned(ptr %base) {
; RV64ZVE32F-NEXT: lbu a2, 0(a2)
; RV64ZVE32F-NEXT: slli a3, a3, 8
; RV64ZVE32F-NEXT: or a2, a3, a2
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e16, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vsetivli zero, 2, e16, m1, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vslideup.vi v8, v9, 1
; RV64ZVE32F-NEXT: andi a2, a1, 4
; RV64ZVE32F-NEXT: beqz a2, .LBB107_3
@@ -13616,9 +13543,8 @@ define <8 x i16> @mgather_strided_2xSEW(ptr %base) {
; RV64ZVE32F-NEXT: .LBB108_10: # %cond.load1
; RV64ZVE32F-NEXT: addi a2, a0, 2
; RV64ZVE32F-NEXT: lh a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e16, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vsetivli zero, 2, e16, m1, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vslideup.vi v8, v9, 1
; RV64ZVE32F-NEXT: andi a2, a1, 4
; RV64ZVE32F-NEXT: beqz a2, .LBB108_3
@@ -13732,9 +13658,8 @@ define <8 x i16> @mgather_strided_2xSEW_with_offset(ptr %base) {
; RV64ZVE32F-NEXT: .LBB109_10: # %cond.load1
; RV64ZVE32F-NEXT: addi a2, a0, 6
; RV64ZVE32F-NEXT: lh a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e16, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vsetivli zero, 2, e16, m1, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vslideup.vi v8, v9, 1
; RV64ZVE32F-NEXT: andi a2, a1, 4
; RV64ZVE32F-NEXT: beqz a2, .LBB109_3
@@ -13848,9 +13773,8 @@ define <8 x i16> @mgather_reverse_unit_strided_2xSEW(ptr %base) {
; RV64ZVE32F-NEXT: .LBB110_10: # %cond.load1
; RV64ZVE32F-NEXT: addi a2, a0, 30
; RV64ZVE32F-NEXT: lh a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e16, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vsetivli zero, 2, e16, m1, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vslideup.vi v8, v9, 1
; RV64ZVE32F-NEXT: andi a2, a1, 4
; RV64ZVE32F-NEXT: beqz a2, .LBB110_3
@@ -13964,9 +13888,8 @@ define <8 x i16> @mgather_reverse_strided_2xSEW(ptr %base) {
; RV64ZVE32F-NEXT: .LBB111_10: # %cond.load1
; RV64ZVE32F-NEXT: addi a2, a0, 30
; RV64ZVE32F-NEXT: lh a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e16, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vsetivli zero, 2, e16, m1, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vslideup.vi v8, v9, 1
; RV64ZVE32F-NEXT: andi a2, a1, 4
; RV64ZVE32F-NEXT: beqz a2, .LBB111_3
@@ -14078,9 +14001,8 @@ define <8 x i16> @mgather_gather_2xSEW(ptr %base) {
; RV64ZVE32F-NEXT: .LBB112_10: # %cond.load1
; RV64ZVE32F-NEXT: addi a2, a0, 2
; RV64ZVE32F-NEXT: lh a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e16, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vsetivli zero, 2, e16, m1, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vslideup.vi v8, v9, 1
; RV64ZVE32F-NEXT: andi a2, a1, 4
; RV64ZVE32F-NEXT: beqz a2, .LBB112_3
@@ -14195,9 +14117,8 @@ define <8 x i16> @mgather_gather_2xSEW_unaligned(ptr %base) {
; RV64ZVE32F-NEXT: .LBB113_10: # %cond.load1
; RV64ZVE32F-NEXT: addi a2, a0, 2
; RV64ZVE32F-NEXT: lh a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e16, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vsetivli zero, 2, e16, m1, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vslideup.vi v8, v9, 1
; RV64ZVE32F-NEXT: andi a2, a1, 4
; RV64ZVE32F-NEXT: beqz a2, .LBB113_3
@@ -14312,9 +14233,8 @@ define <8 x i16> @mgather_gather_2xSEW_unaligned2(ptr %base) {
; RV64ZVE32F-NEXT: ret
; RV64ZVE32F-NEXT: .LBB114_10: # %cond.load1
; RV64ZVE32F-NEXT: lh a3, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e16, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v9, a3
; RV64ZVE32F-NEXT: vsetivli zero, 2, e16, m1, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v9, a3
; RV64ZVE32F-NEXT: vslideup.vi v8, v9, 1
; RV64ZVE32F-NEXT: andi a3, a1, 4
; RV64ZVE32F-NEXT: beqz a3, .LBB114_4
@@ -14432,9 +14352,8 @@ define <8 x i16> @mgather_gather_4xSEW(ptr %base) {
; RV64ZVE32F-NEXT: .LBB115_10: # %cond.load1
; RV64ZVE32F-NEXT: addi a2, a0, 2
; RV64ZVE32F-NEXT: lh a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e16, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vsetivli zero, 2, e16, m1, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vslideup.vi v8, v9, 1
; RV64ZVE32F-NEXT: andi a2, a1, 4
; RV64ZVE32F-NEXT: beqz a2, .LBB115_3
@@ -14550,9 +14469,8 @@ define <8 x i16> @mgather_gather_4xSEW_partial_align(ptr %base) {
; RV64ZVE32F-NEXT: .LBB116_10: # %cond.load1
; RV64ZVE32F-NEXT: addi a2, a0, 2
; RV64ZVE32F-NEXT: lh a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e16, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vsetivli zero, 2, e16, m1, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vslideup.vi v8, v9, 1
; RV64ZVE32F-NEXT: andi a2, a1, 4
; RV64ZVE32F-NEXT: beqz a2, .LBB116_3
@@ -14680,9 +14598,8 @@ define <8 x i16> @mgather_shuffle_rotate(ptr %base) {
; RV64ZVE32F-NEXT: .LBB118_10: # %cond.load1
; RV64ZVE32F-NEXT: addi a2, a0, 10
; RV64ZVE32F-NEXT: lh a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e16, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vsetivli zero, 2, e16, m1, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vslideup.vi v8, v9, 1
; RV64ZVE32F-NEXT: andi a2, a1, 4
; RV64ZVE32F-NEXT: beqz a2, .LBB118_3
@@ -14797,9 +14714,8 @@ define <8 x i16> @mgather_shuffle_vrgather(ptr %base) {
; RV64ZVE32F-NEXT: .LBB119_10: # %cond.load1
; RV64ZVE32F-NEXT: addi a2, a0, 4
; RV64ZVE32F-NEXT: lh a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetvli zero, zero, e16, m1, ta, ma
-; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vsetivli zero, 2, e16, m1, tu, ma
+; RV64ZVE32F-NEXT: vmv.s.x v9, a2
; RV64ZVE32F-NEXT: vslideup.vi v8, v9, 1
; RV64ZVE32F-NEXT: andi a2, a1, 4
; RV64ZVE32F-NEXT: beqz a2, .LBB119_3
diff --git a/llvm/test/CodeGen/RISCV/rvv/vsetvli-insert-crossbb.ll b/llvm/test/CodeGen/RISCV/rvv/vsetvli-insert-crossbb.ll
index 73f651225da64d3..ce7a284bb6f9a66 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vsetvli-insert-crossbb.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vsetvli-insert-crossbb.ll
@@ -833,8 +833,7 @@ define <vscale x 2 x i32> @pre_lmul(<vscale x 2 x i32> %x, <vscale x 2 x i32> %y
; CHECK-LABEL: pre_lmul:
; CHECK: # %bb.0: # %entry
; CHECK-NEXT: andi a0, a0, 1
-; CHECK-NEXT: vsetvli a1, zero, e64, m1, ta, ma
-; CHECK-NEXT: vsetvli zero, a1, e32, m1, ta, ma
+; CHECK-NEXT: vsetvli a1, zero, e32, m1, ta, ma
; CHECK-NEXT: vadd.vv v8, v8, v9
; CHECK-NEXT: vadd.vv v8, v8, v9
; CHECK-NEXT: ret
diff --git a/llvm/test/CodeGen/RISCV/rvv/vsetvli-insert.ll b/llvm/test/CodeGen/RISCV/rvv/vsetvli-insert.ll
index 4d0f640408dd2a9..1e2049e3e8a4d52 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vsetvli-insert.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vsetvli-insert.ll
@@ -329,9 +329,9 @@ entry:
define double @test17(i64 %avl, <vscale x 1 x double> %a, <vscale x 1 x double> %b) nounwind {
; CHECK-LABEL: test17:
; CHECK: # %bb.0: # %entry
-; CHECK-NEXT: vsetvli a0, a0, e64, m1, ta, ma
-; CHECK-NEXT: vfmv.f.s fa5, v8
+; CHECK-NEXT: vsetvli a0, a0, e32, mf2, ta, ma
; CHECK-NEXT: vsetvli zero, a0, e64, m1, ta, ma
+; CHECK-NEXT: vfmv.f.s fa5, v8
; CHECK-NEXT: vfadd.vv v8, v8, v9
; CHECK-NEXT: vfmv.f.s fa4, v8
; CHECK-NEXT: fadd.d fa0, fa5, fa4
@@ -468,7 +468,6 @@ entry:
define void @avl_forward4(<vscale x 2 x i32> %v, <vscale x 2 x i32>* %p, i64 %reg) nounwind {
; CHECK-LABEL: avl_forward4:
; CHECK: # %bb.0: # %entry
-; CHECK-NEXT: vsetvli a1, a1, e16, m1, ta, ma
; CHECK-NEXT: vsetvli zero, a1, e32, m1, ta, ma
; CHECK-NEXT: vse32.v v8, (a0)
; CHECK-NEXT: ret
@@ -498,7 +497,6 @@ entry:
define <vscale x 1 x i64> @vleNff(i64* %str, i64 %n, i64 %x) {
; CHECK-LABEL: vleNff:
; CHECK: # %bb.0: # %entry
-; CHECK-NEXT: vsetvli a1, a1, e8, m4, ta, ma
; CHECK-NEXT: vsetvli zero, a1, e64, m1, ta, ma
; CHECK-NEXT: vle64ff.v v8, (a0)
; CHECK-NEXT: vsetvli zero, zero, e64, m1, tu, ma
@@ -519,7 +517,6 @@ entry:
define <vscale x 1 x i64> @vleNff2(i64* %str, i64 %n, i64 %x) {
; CHECK-LABEL: vleNff2:
; CHECK: # %bb.0: # %entry
-; CHECK-NEXT: vsetvli a1, a1, e8, m4, ta, ma
; CHECK-NEXT: vsetvli zero, a1, e64, m1, ta, ma
; CHECK-NEXT: vle64ff.v v8, (a0)
; CHECK-NEXT: vadd.vx v8, v8, a2
@@ -545,7 +542,6 @@ define <vscale x 2 x i32> @avl_forward5(<vscale x 2 x i32>* %addr) {
; CHECK-LABEL: avl_forward5:
; CHECK: # %bb.0:
; CHECK-NEXT: li a1, 32
-; CHECK-NEXT: vsetvli a1, a1, e8, m4, ta, ma
; CHECK-NEXT: vsetvli zero, a1, e32, m1, ta, ma
; CHECK-NEXT: vle32.v v8, (a0)
; CHECK-NEXT: ret
diff --git a/llvm/test/CodeGen/RISCV/rvv/vsetvli-intrinsics.ll b/llvm/test/CodeGen/RISCV/rvv/vsetvli-intrinsics.ll
index 11b164fbf51e830..00917d05b42e8cf 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vsetvli-intrinsics.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vsetvli-intrinsics.ll
@@ -118,12 +118,9 @@ define <vscale x 4 x i32> @redundant_vsetvli(iXLen %avl, <vscale x 4 x i32>* %pt
; Check that we remove the repeated/redundant vsetvli when followed by another
; operation
-; FIXME: We don't catch the second vsetvli because it has a use of its output.
-; We could replace it with the output of the first vsetvli.
define <vscale x 4 x i32> @repeated_vsetvli(iXLen %avl, <vscale x 4 x i32>* %ptr) nounwind {
; CHECK-LABEL: repeated_vsetvli:
; CHECK: # %bb.0:
-; CHECK-NEXT: vsetvli a0, a0, e32, m2, ta, ma
; CHECK-NEXT: vsetvli zero, a0, e32, m2, ta, ma
; CHECK-NEXT: vle32.v v8, (a1)
; CHECK-NEXT: ret
@@ -156,3 +153,18 @@ define iXLen @test_vsetvli_eqvlmax_e8m8(iXLen %avl) nounwind {
%vl = call iXLen @llvm.riscv.vsetvli.iXLen(iXLen 128, iXLen 0, iXLen 3)
ret iXLen %vl
}
+
+; Check that we remove the repeated/redundant vsetvli, even in the case when
+; their results may be used in subsequent vsetvli (but nowhere else)
+define <vscale x 4 x i32> @chained_vsetvli(iXLen %avl, <vscale x 4 x i32>* %ptr) nounwind {
+; CHECK-LABEL: chained_vsetvli:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vsetvli zero, a0, e32, m2, ta, ma
+; CHECK-NEXT: vle32.v v8, (a1)
+; CHECK-NEXT: ret
+ %vl0 = call iXLen @llvm.riscv.vsetvli.iXLen(iXLen %avl, iXLen 2, iXLen 1)
+ %vl1 = call iXLen @llvm.riscv.vsetvli.iXLen(iXLen %vl0, iXLen 2, iXLen 1)
+ %vl2 = call iXLen @llvm.riscv.vsetvli.iXLen(iXLen %vl1, iXLen 2, iXLen 1)
+ %x = call <vscale x 4 x i32> @llvm.riscv.vle.nxv4i32.iXLen(<vscale x 4 x i32> undef, <vscale x 4 x i32>* %ptr, iXLen %vl1)
+ ret <vscale x 4 x i32> %x
+}
>From 15e2a7f1b4975fffcde9c708f92b2d839a3be2b3 Mon Sep 17 00:00:00 2001
From: Simeon Krastnikov <simeon.krastnikov at imgtec.com>
Date: Mon, 25 Sep 2023 10:14:48 +0100
Subject: [PATCH 2/2] fixup! Improve cleanup phase of RISCV Insert VSETVLI pass
---
llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp | 18 +++++++++++++-----
1 file changed, 13 insertions(+), 5 deletions(-)
diff --git a/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp b/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp
index 138c28e4effabf6..620009d307be29f 100644
--- a/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp
+++ b/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp
@@ -1438,11 +1438,19 @@ void RISCVInsertVSETVLI::doLocalPostpass(MachineBasicBlock &MBB) {
if (!isVectorConfigInstr(MI)) {
doUnion(Used, getDemanded(MI, MRI));
- // We can't handle the case when the source AVL
- // register of *NextMI is defined after MI
- if (NextMI && NextMI->getOperand(1).isReg() && MI.getOperand(0).isReg() &&
- MI.getOperand(0).getReg() == NextMI->getOperand(1).getReg())
- Used.demandVL();
+
+ // We can't handle the case when the source AVL register of *NextMI is
+ // defined after MI
+ if (NextMI && NextMI->getOperand(1).isReg()) {
+ for (const MachineOperand &MO : MI.operands()) {
+ if (MO.isReg() && MO.isDef() &&
+ MO.getReg() == NextMI->getOperand(1).getReg()) {
+ Used.demandVL();
+ break;
+ }
+ }
+ }
+
continue;
}
More information about the llvm-commits
mailing list