[llvm] [RISCV] Combine vslideup_vl with known VL to a smaller LMUL (PR #66671)
via llvm-commits
llvm-commits at lists.llvm.org
Mon Sep 18 10:01:47 PDT 2023
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-risc-v
<details>
<summary>Changes</summary>
Similiar to #<!-- -->66267, we can perform a vslideup_vl on a smaller type if we know the highest lane that will be written to, which can be determined from VL.
This is an alternative to #<!-- -->65997 and #<!-- -->66087
Stacked upon #<!-- -->66267
---
Patch is 127.17 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/66671.diff
13 Files Affected:
- (modified) llvm/lib/Target/RISCV/RISCVISelLowering.cpp (+47-9)
- (modified) llvm/test/CodeGen/RISCV/rvv/extractelt-int-rv32.ll (+10-8)
- (modified) llvm/test/CodeGen/RISCV/rvv/fixed-vectors-insert-i1.ll (+1-1)
- (modified) llvm/test/CodeGen/RISCV/rvv/fixed-vectors-insert-subvector.ll (+6-6)
- (modified) llvm/test/CodeGen/RISCV/rvv/fixed-vectors-insert.ll (+1-1)
- (modified) llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-explodevector.ll (+102-80)
- (modified) llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll (+129-129)
- (modified) llvm/test/CodeGen/RISCV/rvv/fixed-vectors-strided-load-combine.ll (+6-6)
- (modified) llvm/test/CodeGen/RISCV/rvv/fpclamptosat_vec.ll (+435-213)
- (modified) llvm/test/CodeGen/RISCV/rvv/insertelt-fp.ll (+8-8)
- (modified) llvm/test/CodeGen/RISCV/rvv/insertelt-i1.ll (+3-3)
- (modified) llvm/test/CodeGen/RISCV/rvv/insertelt-int-rv32.ll (+11-11)
- (modified) llvm/test/CodeGen/RISCV/rvv/insertelt-int-rv64.ll (+11-11)
``````````diff
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index de58335b435651c..179db28ec83e476 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -8805,15 +8805,6 @@ SDValue RISCVTargetLowering::lowerEXTRACT_SUBVECTOR(SDValue Op,
Vec = convertToScalableVector(ContainerVT, Vec, DAG, Subtarget);
}
- // Shrink down Vec so we're performing the slidedown on a smaller LMUL.
- unsigned LastIdx = OrigIdx + SubVecVT.getVectorNumElements() - 1;
- if (auto ShrunkVT =
- getSmallestVTForIndex(ContainerVT, LastIdx, DL, DAG, Subtarget)) {
- ContainerVT = *ShrunkVT;
- Vec = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, ContainerVT, Vec,
- DAG.getVectorIdxConstant(0, DL));
- }
-
SDValue Mask =
getDefaultVLOps(VecVT, ContainerVT, DL, DAG, Subtarget).first;
// Set the vector length to only the number of elements we care about. This
@@ -14260,6 +14251,53 @@ SDValue RISCVTargetLowering::PerformDAGCombine(SDNode *N,
if (SDValue V = performCONCAT_VECTORSCombine(N, DAG, Subtarget, *this))
return V;
break;
+ case RISCVISD::VSLIDEUP_VL:
+ case RISCVISD::VSLIDEDOWN_VL: {
+ MVT OrigVT = N->getSimpleValueType(0);
+ auto *CVL = dyn_cast<ConstantSDNode>(N->getOperand(4));
+ if (!CVL)
+ break;
+
+ // The maximum index read or written is VL - 1 for vslideup, and VL + offset
+ // - 1 for vslidedown.
+ unsigned MaxIdx = CVL->getZExtValue() - 1;
+ if (N->getOpcode() == RISCVISD::VSLIDEDOWN_VL) {
+ auto *COffset = dyn_cast<ConstantSDNode>(N->getOperand(2));
+ if (!COffset)
+ break;
+ MaxIdx += COffset->getZExtValue();
+ }
+
+ // We can try and reduce the LMUL that a vslide* uses if we know where
+ // the maximum index is. For example, if the target has Zvl128b, a
+ // vslidedown of e32 with with an offset of 4 and VL of 2 is only going to
+ // read from the first 2 registers at most. So if we were operating at
+ // LMUL=4 (nxv8i32), we can reduce it to LMUL=2(nxv4i32).
+ if (auto ShrunkVT =
+ getSmallestVTForIndex(OrigVT, MaxIdx, DL, DAG, Subtarget)) {
+ SDValue ShrunkPassthru =
+ DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, *ShrunkVT, N->getOperand(0),
+ DAG.getVectorIdxConstant(0, DL));
+ SDValue ShrunkInVec =
+ DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, *ShrunkVT, N->getOperand(1),
+ DAG.getVectorIdxConstant(0, DL));
+
+ // The only mask ever used in vslide*_vl nodes is vmset_vl, and the only
+ // patterns on vslide*_vl only accept vmset_vl. So create a new vmset
+ // since using an extract_subvector breaks patterns.
+ assert(N->getOperand(3).getOpcode() == RISCVISD::VMSET_VL);
+ SDValue ShrunkMask =
+ DAG.getNode(RISCVISD::VMSET_VL, SDLoc(N), getMaskTypeFor(*ShrunkVT),
+ N->getOperand(4));
+ SDValue ShrunkSlidedown =
+ DAG.getNode(N->getOpcode(), DL, *ShrunkVT,
+ {ShrunkPassthru, ShrunkInVec, N->getOperand(2),
+ ShrunkMask, N->getOperand(4), N->getOperand(5)});
+ return DAG.getNode(ISD::INSERT_SUBVECTOR, DL, OrigVT, N->getOperand(0),
+ ShrunkSlidedown, DAG.getVectorIdxConstant(0, DL));
+ }
+ break;
+ }
case RISCVISD::VFMV_V_F_VL: {
const MVT VT = N->getSimpleValueType(0);
SDValue Passthru = N->getOperand(0);
diff --git a/llvm/test/CodeGen/RISCV/rvv/extractelt-int-rv32.ll b/llvm/test/CodeGen/RISCV/rvv/extractelt-int-rv32.ll
index fd2f89e26e59809..c3181a296abe06d 100644
--- a/llvm/test/CodeGen/RISCV/rvv/extractelt-int-rv32.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/extractelt-int-rv32.ll
@@ -679,12 +679,13 @@ define i64 @extractelt_nxv4i64_0(<vscale x 4 x i64> %v) {
define i64 @extractelt_nxv4i64_imm(<vscale x 4 x i64> %v) {
; CHECK-LABEL: extractelt_nxv4i64_imm:
; CHECK: # %bb.0:
-; CHECK-NEXT: vsetivli zero, 1, e64, m4, ta, ma
+; CHECK-NEXT: vsetivli zero, 1, e64, m2, ta, ma
; CHECK-NEXT: vslidedown.vi v8, v8, 2
-; CHECK-NEXT: li a0, 32
-; CHECK-NEXT: vsrl.vx v12, v8, a0
-; CHECK-NEXT: vmv.x.s a1, v12
; CHECK-NEXT: vmv.x.s a0, v8
+; CHECK-NEXT: li a1, 32
+; CHECK-NEXT: vsetivli zero, 1, e64, m4, ta, ma
+; CHECK-NEXT: vsrl.vx v8, v8, a1
+; CHECK-NEXT: vmv.x.s a1, v8
; CHECK-NEXT: ret
%r = extractelement <vscale x 4 x i64> %v, i32 2
ret i64 %r
@@ -720,12 +721,13 @@ define i64 @extractelt_nxv8i64_0(<vscale x 8 x i64> %v) {
define i64 @extractelt_nxv8i64_imm(<vscale x 8 x i64> %v) {
; CHECK-LABEL: extractelt_nxv8i64_imm:
; CHECK: # %bb.0:
-; CHECK-NEXT: vsetivli zero, 1, e64, m8, ta, ma
+; CHECK-NEXT: vsetivli zero, 1, e64, m2, ta, ma
; CHECK-NEXT: vslidedown.vi v8, v8, 2
-; CHECK-NEXT: li a0, 32
-; CHECK-NEXT: vsrl.vx v16, v8, a0
-; CHECK-NEXT: vmv.x.s a1, v16
; CHECK-NEXT: vmv.x.s a0, v8
+; CHECK-NEXT: li a1, 32
+; CHECK-NEXT: vsetivli zero, 1, e64, m8, ta, ma
+; CHECK-NEXT: vsrl.vx v8, v8, a1
+; CHECK-NEXT: vmv.x.s a1, v8
; CHECK-NEXT: ret
%r = extractelement <vscale x 8 x i64> %v, i32 2
ret i64 %r
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-insert-i1.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-insert-i1.ll
index b3cbad3d9e6b1d7..f7737784d4ca57e 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-insert-i1.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-insert-i1.ll
@@ -108,7 +108,7 @@ define <64 x i1> @insertelt_v64i1(<64 x i1> %x, i1 %elt) nounwind {
; CHECK-NEXT: vmv.v.i v8, 0
; CHECK-NEXT: vmerge.vim v8, v8, 1, v0
; CHECK-NEXT: vmv.s.x v12, a0
-; CHECK-NEXT: vsetivli zero, 2, e8, m4, tu, ma
+; CHECK-NEXT: vsetivli zero, 2, e8, m1, tu, ma
; CHECK-NEXT: vslideup.vi v8, v12, 1
; CHECK-NEXT: vsetvli zero, a1, e8, m4, ta, ma
; CHECK-NEXT: vand.vi v8, v8, 1
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-insert-subvector.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-insert-subvector.ll
index 1d6a45ed36f335c..133b09428ed961b 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-insert-subvector.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-insert-subvector.ll
@@ -27,7 +27,7 @@ define <vscale x 8 x i32> @insert_nxv8i32_v2i32_2(<vscale x 8 x i32> %vec, ptr %
; CHECK: # %bb.0:
; CHECK-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
; CHECK-NEXT: vle32.v v12, (a0)
-; CHECK-NEXT: vsetivli zero, 4, e32, m4, tu, ma
+; CHECK-NEXT: vsetivli zero, 4, e32, m1, tu, ma
; CHECK-NEXT: vslideup.vi v8, v12, 2
; CHECK-NEXT: ret
%sv = load <2 x i32>, ptr %svp
@@ -40,7 +40,7 @@ define <vscale x 8 x i32> @insert_nxv8i32_v2i32_6(<vscale x 8 x i32> %vec, ptr %
; CHECK: # %bb.0:
; CHECK-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
; CHECK-NEXT: vle32.v v12, (a0)
-; CHECK-NEXT: vsetivli zero, 8, e32, m4, tu, ma
+; CHECK-NEXT: vsetivli zero, 8, e32, m2, tu, ma
; CHECK-NEXT: vslideup.vi v8, v12, 6
; CHECK-NEXT: ret
%sv = load <2 x i32>, ptr %svp
@@ -65,7 +65,7 @@ define <vscale x 8 x i32> @insert_nxv8i32_v8i32_0(<vscale x 8 x i32> %vec, ptr %
; LMULMAX1-NEXT: vle32.v v16, (a0)
; LMULMAX1-NEXT: vsetivli zero, 4, e32, m4, tu, ma
; LMULMAX1-NEXT: vmv.v.v v8, v12
-; LMULMAX1-NEXT: vsetivli zero, 8, e32, m4, tu, ma
+; LMULMAX1-NEXT: vsetivli zero, 8, e32, m2, tu, ma
; LMULMAX1-NEXT: vslideup.vi v8, v16, 4
; LMULMAX1-NEXT: ret
%sv = load <8 x i32>, ptr %svp
@@ -197,7 +197,7 @@ define void @insert_v8i32_v2i32_2(ptr %vp, ptr %svp) {
; LMULMAX2-NEXT: vle32.v v8, (a1)
; LMULMAX2-NEXT: vsetivli zero, 8, e32, m2, ta, ma
; LMULMAX2-NEXT: vle32.v v10, (a0)
-; LMULMAX2-NEXT: vsetivli zero, 4, e32, m2, tu, ma
+; LMULMAX2-NEXT: vsetivli zero, 4, e32, m1, tu, ma
; LMULMAX2-NEXT: vslideup.vi v10, v8, 2
; LMULMAX2-NEXT: vsetivli zero, 8, e32, m2, ta, ma
; LMULMAX2-NEXT: vse32.v v10, (a0)
@@ -509,7 +509,7 @@ define void @insert_v2i64_nxv16i64(ptr %psv0, ptr %psv1, <vscale x 16 x i64>* %o
; CHECK-NEXT: vsetivli zero, 2, e64, m1, ta, ma
; CHECK-NEXT: vle64.v v8, (a0)
; CHECK-NEXT: vle64.v v16, (a1)
-; CHECK-NEXT: vsetivli zero, 6, e64, m8, tu, ma
+; CHECK-NEXT: vsetivli zero, 6, e64, m4, tu, ma
; CHECK-NEXT: vslideup.vi v8, v16, 4
; CHECK-NEXT: vs8r.v v8, (a2)
; CHECK-NEXT: ret
@@ -539,7 +539,7 @@ define void @insert_v2i64_nxv16i64_lo2(ptr %psv, <vscale x 16 x i64>* %out) {
; CHECK: # %bb.0:
; CHECK-NEXT: vsetivli zero, 2, e64, m1, ta, ma
; CHECK-NEXT: vle64.v v8, (a0)
-; CHECK-NEXT: vsetivli zero, 4, e64, m8, ta, ma
+; CHECK-NEXT: vsetivli zero, 4, e64, m2, ta, ma
; CHECK-NEXT: vslideup.vi v16, v8, 2
; CHECK-NEXT: vs8r.v v16, (a1)
; CHECK-NEXT: ret
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-insert.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-insert.ll
index 373a96356a207e2..6f5ab60fb4ad003 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-insert.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-insert.ll
@@ -54,7 +54,7 @@ define <32 x i32> @insertelt_v32i32_4(<32 x i32> %a, i32 %y) {
; CHECK-NEXT: li a1, 32
; CHECK-NEXT: vsetvli zero, a1, e32, m1, ta, ma
; CHECK-NEXT: vmv.s.x v16, a0
-; CHECK-NEXT: vsetivli zero, 5, e32, m8, tu, ma
+; CHECK-NEXT: vsetivli zero, 5, e32, m2, tu, ma
; CHECK-NEXT: vslideup.vi v8, v16, 4
; CHECK-NEXT: ret
%b = insertelement <32 x i32> %a, i32 %y, i32 4
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-explodevector.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-explodevector.ll
index 4e60edf058450f0..eb74c5e608b93ee 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-explodevector.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-explodevector.ll
@@ -811,9 +811,11 @@ define i64 @explode_4xi64(<4 x i64> %v) {
; RV32-NEXT: vsrl.vx v10, v8, a0
; RV32-NEXT: vmv.x.s a1, v10
; RV32-NEXT: vmv.x.s a2, v8
+; RV32-NEXT: vsetivli zero, 1, e64, m1, ta, ma
; RV32-NEXT: vslidedown.vi v10, v8, 1
-; RV32-NEXT: vsrl.vx v12, v10, a0
-; RV32-NEXT: vmv.x.s a3, v12
+; RV32-NEXT: vmv.x.s a3, v10
+; RV32-NEXT: vsetivli zero, 1, e64, m2, ta, ma
+; RV32-NEXT: vsrl.vx v10, v10, a0
; RV32-NEXT: vmv.x.s a4, v10
; RV32-NEXT: vslidedown.vi v10, v8, 2
; RV32-NEXT: vsrl.vx v12, v10, a0
@@ -823,12 +825,12 @@ define i64 @explode_4xi64(<4 x i64> %v) {
; RV32-NEXT: vsrl.vx v10, v8, a0
; RV32-NEXT: vmv.x.s a0, v10
; RV32-NEXT: vmv.x.s a7, v8
-; RV32-NEXT: add a1, a1, a3
-; RV32-NEXT: add a4, a2, a4
-; RV32-NEXT: sltu a2, a4, a2
+; RV32-NEXT: add a1, a1, a4
+; RV32-NEXT: add a3, a2, a3
+; RV32-NEXT: sltu a2, a3, a2
; RV32-NEXT: add a1, a1, a2
-; RV32-NEXT: add a6, a4, a6
-; RV32-NEXT: sltu a2, a6, a4
+; RV32-NEXT: add a6, a3, a6
+; RV32-NEXT: sltu a2, a6, a3
; RV32-NEXT: add a1, a1, a5
; RV32-NEXT: add a0, a2, a0
; RV32-NEXT: add a1, a1, a0
@@ -875,15 +877,21 @@ define i64 @explode_8xi64(<8 x i64> %v) {
; RV32-NEXT: vsrl.vx v12, v8, a0
; RV32-NEXT: vmv.x.s a1, v12
; RV32-NEXT: vmv.x.s a2, v8
+; RV32-NEXT: vsetivli zero, 1, e64, m1, ta, ma
; RV32-NEXT: vslidedown.vi v12, v8, 1
-; RV32-NEXT: vsrl.vx v16, v12, a0
-; RV32-NEXT: vmv.x.s a3, v16
+; RV32-NEXT: vmv.x.s a3, v12
+; RV32-NEXT: vsetivli zero, 1, e64, m4, ta, ma
+; RV32-NEXT: vsrl.vx v12, v12, a0
; RV32-NEXT: vmv.x.s a4, v12
+; RV32-NEXT: vsetivli zero, 1, e64, m2, ta, ma
; RV32-NEXT: vslidedown.vi v12, v8, 2
+; RV32-NEXT: vsetivli zero, 1, e64, m4, ta, ma
; RV32-NEXT: vsrl.vx v16, v12, a0
; RV32-NEXT: vmv.x.s a5, v16
; RV32-NEXT: vmv.x.s a6, v12
+; RV32-NEXT: vsetivli zero, 1, e64, m2, ta, ma
; RV32-NEXT: vslidedown.vi v12, v8, 3
+; RV32-NEXT: vsetivli zero, 1, e64, m4, ta, ma
; RV32-NEXT: vsrl.vx v16, v12, a0
; RV32-NEXT: vmv.x.s a7, v16
; RV32-NEXT: vmv.x.s t0, v12
@@ -903,19 +911,19 @@ define i64 @explode_8xi64(<8 x i64> %v) {
; RV32-NEXT: vsrl.vx v12, v8, a0
; RV32-NEXT: vmv.x.s a0, v12
; RV32-NEXT: vmv.x.s s0, v8
-; RV32-NEXT: add a1, a1, a3
-; RV32-NEXT: add a4, a2, a4
-; RV32-NEXT: sltu a2, a4, a2
+; RV32-NEXT: add a1, a1, a4
+; RV32-NEXT: add a3, a2, a3
+; RV32-NEXT: sltu a2, a3, a2
; RV32-NEXT: add a1, a1, a2
-; RV32-NEXT: add a6, a4, a6
-; RV32-NEXT: sltu a2, a6, a4
+; RV32-NEXT: add a6, a3, a6
+; RV32-NEXT: sltu a2, a6, a3
; RV32-NEXT: add a1, a1, a5
-; RV32-NEXT: add a2, a2, a7
-; RV32-NEXT: add a1, a1, a2
; RV32-NEXT: add t0, a6, t0
-; RV32-NEXT: sltu a2, t0, a6
-; RV32-NEXT: add a2, a2, t1
+; RV32-NEXT: sltu a3, t0, a6
+; RV32-NEXT: add a2, a2, a7
; RV32-NEXT: add a1, a1, a2
+; RV32-NEXT: add a3, a3, t1
+; RV32-NEXT: add a1, a1, a3
; RV32-NEXT: add t2, t0, t2
; RV32-NEXT: sltu a2, t2, t0
; RV32-NEXT: add a2, a2, t3
@@ -1029,115 +1037,129 @@ define i64 @explode_16xi64(<16 x i64> %v) {
; RV32-NEXT: vmv.x.s a0, v16
; RV32-NEXT: sw a0, 8(sp) # 4-byte Folded Spill
; RV32-NEXT: vmv.x.s a0, v8
+; RV32-NEXT: vsetivli zero, 1, e64, m1, ta, ma
; RV32-NEXT: vslidedown.vi v16, v8, 1
+; RV32-NEXT: vmv.x.s a3, v16
+; RV32-NEXT: vsetivli zero, 1, e64, m8, ta, ma
+; RV32-NEXT: vsrl.vx v16, v16, a1
+; RV32-NEXT: vmv.x.s a4, v16
+; RV32-NEXT: vsetivli zero, 1, e64, m2, ta, ma
+; RV32-NEXT: vslidedown.vi v16, v8, 2
+; RV32-NEXT: vsetivli zero, 1, e64, m8, ta, ma
; RV32-NEXT: vsrl.vx v24, v16, a1
; RV32-NEXT: vmv.x.s a5, v24
; RV32-NEXT: vmv.x.s a6, v16
-; RV32-NEXT: vslidedown.vi v16, v8, 2
-; RV32-NEXT: vsrl.vx v24, v16, a1
-; RV32-NEXT: vmv.x.s a3, v24
-; RV32-NEXT: vmv.x.s a4, v16
+; RV32-NEXT: vsetivli zero, 1, e64, m2, ta, ma
; RV32-NEXT: vslidedown.vi v16, v8, 3
+; RV32-NEXT: vsetivli zero, 1, e64, m8, ta, ma
; RV32-NEXT: vsrl.vx v24, v16, a1
-; RV32-NEXT: vmv.x.s s2, v24
+; RV32-NEXT: vmv.x.s t0, v24
; RV32-NEXT: vmv.x.s a7, v16
+; RV32-NEXT: vsetivli zero, 1, e64, m4, ta, ma
; RV32-NEXT: vslidedown.vi v16, v8, 4
-; RV32-NEXT: vsrl.vx v24, v16, a1
-; RV32-NEXT: vmv.x.s s3, v24
-; RV32-NEXT: vmv.x.s t0, v16
-; RV32-NEXT: vslidedown.vi v16, v8, 5
-; RV32-NEXT: vsrl.vx v24, v16, a1
-; RV32-NEXT: vmv.x.s s4, v24
; RV32-NEXT: vmv.x.s t1, v16
-; RV32-NEXT: vslidedown.vi v16, v8, 6
-; RV32-NEXT: vsrl.vx v24, v16, a1
-; RV32-NEXT: vmv.x.s s5, v24
+; RV32-NEXT: vsetivli zero, 1, e64, m8, ta, ma
+; RV32-NEXT: vsrl.vx v16, v16, a1
+; RV32-NEXT: vmv.x.s t3, v16
+; RV32-NEXT: vsetivli zero, 1, e64, m4, ta, ma
+; RV32-NEXT: vslidedown.vi v16, v8, 5
; RV32-NEXT: vmv.x.s t2, v16
+; RV32-NEXT: vsetivli zero, 1, e64, m8, ta, ma
+; RV32-NEXT: vsrl.vx v16, v16, a1
+; RV32-NEXT: vmv.x.s t5, v16
+; RV32-NEXT: vsetivli zero, 1, e64, m4, ta, ma
+; RV32-NEXT: vslidedown.vi v16, v8, 6
+; RV32-NEXT: vmv.x.s t4, v16
+; RV32-NEXT: vsetivli zero, 1, e64, m8, ta, ma
+; RV32-NEXT: vsrl.vx v16, v16, a1
+; RV32-NEXT: vmv.x.s s0, v16
+; RV32-NEXT: vsetivli zero, 1, e64, m4, ta, ma
; RV32-NEXT: vslidedown.vi v16, v8, 7
-; RV32-NEXT: vsrl.vx v24, v16, a1
-; RV32-NEXT: vmv.x.s s6, v24
-; RV32-NEXT: vmv.x.s t3, v16
+; RV32-NEXT: vmv.x.s t6, v16
+; RV32-NEXT: vsetivli zero, 1, e64, m8, ta, ma
+; RV32-NEXT: vsrl.vx v16, v16, a1
+; RV32-NEXT: vmv.x.s ra, v16
; RV32-NEXT: vslidedown.vi v16, v8, 8
; RV32-NEXT: vsrl.vx v24, v16, a1
-; RV32-NEXT: vmv.x.s s7, v24
-; RV32-NEXT: vmv.x.s t4, v16
+; RV32-NEXT: vmv.x.s s6, v24
+; RV32-NEXT: vmv.x.s s1, v16
; RV32-NEXT: vslidedown.vi v16, v8, 9
; RV32-NEXT: vsrl.vx v24, v16, a1
-; RV32-NEXT: vmv.x.s s8, v24
-; RV32-NEXT: vmv.x.s t5, v16
+; RV32-NEXT: vmv.x.s s7, v24
+; RV32-NEXT: vmv.x.s s2, v16
; RV32-NEXT: vslidedown.vi v16, v8, 10
; RV32-NEXT: vsrl.vx v24, v16, a1
-; RV32-NEXT: vmv.x.s s9, v24
-; RV32-NEXT: vmv.x.s t6, v16
+; RV32-NEXT: vmv.x.s s8, v24
+; RV32-NEXT: vmv.x.s s3, v16
; RV32-NEXT: vslidedown.vi v16, v8, 11
; RV32-NEXT: vsrl.vx v24, v16, a1
-; RV32-NEXT: vmv.x.s s10, v24
-; RV32-NEXT: vmv.x.s s0, v16
+; RV32-NEXT: vmv.x.s s9, v24
+; RV32-NEXT: vmv.x.s s4, v16
; RV32-NEXT: vslidedown.vi v16, v8, 12
; RV32-NEXT: vsrl.vx v24, v16, a1
-; RV32-NEXT: vmv.x.s s11, v24
-; RV32-NEXT: vmv.x.s s1, v16
+; RV32-NEXT: vmv.x.s s10, v24
+; RV32-NEXT: vmv.x.s s5, v16
; RV32-NEXT: vslidedown.vi v0, v8, 13
; RV32-NEXT: vsrl.vx v16, v0, a1
-; RV32-NEXT: vmv.x.s ra, v16
+; RV32-NEXT: vmv.x.s s11, v16
; RV32-NEXT: vslidedown.vi v16, v8, 14
; RV32-NEXT: vsrl.vx v24, v16, a1
; RV32-NEXT: vslidedown.vi v8, v8, 15
; RV32-NEXT: vmv.x.s a2, v0
; RV32-NEXT: vsrl.vx v0, v8, a1
; RV32-NEXT: lw a1, 8(sp) # 4-byte Folded Reload
-; RV32-NEXT: add a5, a1, a5
-; RV32-NEXT: add a6, a0, a6
-; RV32-NEXT: sltu a0, a6, a0
-; RV32-NEXT: add a0, a5, a0
-; RV32-NEXT: add a0, a0, a3
-; RV32-NEXT: add a4, a6, a4
-; RV32-NEXT: sltu a1, a4, a6
-; RV32-NEXT: add a1, a1, s2
-; RV32-NEXT: add a0, a0, a1
-; RV32-NEXT: add a7, a4, a7
-; RV32-NEXT: sltu a1, a7, a4
-; RV32-NEXT: add a1, a1, s3
+; RV32-NEXT: add a4, a1, a4
+; RV32-NEXT: add a3, a0, a3
+; RV32-NEXT: sltu a0, a3, a0
+; RV32-NEXT: add a0, a4, a0
+; RV32-NEXT: add a0, a0, a5
+; RV32-NEXT: add a6, a3, a6
+; RV32-NEXT: sltu a1, a6, a3
+; RV32-NEXT: add a1, a1, t0
; RV32-NEXT: add a0, a0, a1
-; RV32-NEXT: add t0, a7, t0
-; RV32-NEXT: sltu a1, t0, a7
-; RV32-NEXT: add a1, a1, s4
+; RV32-NEXT: add a7, a6, a7
+; RV32-NEXT: sltu a1, a7, a6
+; RV32-NEXT: add a1, a1, t3
; RV32-NEXT: add a0, a0, a1
-; RV32-NEXT: add t1, t0, t1
-; RV32-NEXT: sltu a1, t1, t0
-; RV32-NEXT: add a1, a1, s5
+; RV32-NEXT: add t1, a7, t1
+; RV32-NEXT: sltu a1, t1, a7
+; RV32-NEXT: add a1, a1, t5
; RV32-NEXT: add a0, a0, a1
; RV32-NEXT: add t2, t1, t2
; RV32-NEXT: sltu a1, t2, t1
+; RV32-NEXT: add a1, a1, s0
+; RV32-NEXT: add a0, a0, a1
+; RV32-NEXT: add t4, t2, t4
+; RV32-NEXT: sltu a1, t4, t2
+; RV32-NEXT: add a1, a1, ra
+; RV32-NEXT: add a0, a0, a1
+; RV32-NEXT: add t6, t4, t6
+; RV32-NEXT: sltu a1, t6, t4
; RV32-NEXT: add a1, a1, s6
; RV32-NEXT: add a0, a0, a1
-; RV32-NEXT: add t3, t2, t3
-; RV32-NEXT: sltu a1, t3, t2
+; RV32-NEXT: add s1, t6, s1
+; RV32-NEXT: sltu a1, s1, t6
; RV32-NEXT: add a1, a1, s7
; RV32-NEXT: add a0, a0, a1
-; RV32-NEXT: add t4, t3, t4
-; RV32-NEXT: sltu a1, t4, t3
+; RV32-NEXT: add s2, s1, s2
+; RV32-NEXT: sltu a1, s2, s1
; RV32-NEXT: add a1, a1, s8
; RV32-NEXT: add a0, a0, a1
-; RV32-NEXT: add t5, t4, t5
-; RV32-NEXT: sltu a1, t5, t4
+; RV32-NEXT: add s3, s2, s3
+; RV32-NEXT: sltu a1, s3, s2
; RV32-NEXT: add a1, a1, s9
; RV32-NEXT: add a0, a0, a1
-; RV32-NEXT: add t6, t5, t6
-; RV32-NEXT: sltu a1, t6, t5
+; RV32-NEXT: add s4, s3, s4
+; RV32-NEXT: sltu a1, s4, s3
; RV32-NEXT: add a1, a1, s10
; RV32-NEXT: add a0, a0, a1
-; RV32-NEXT: add s0, t6, s0
-; RV32-NEXT: sltu a1, s0, t6
+; RV32-NEXT: add s5, s4, s5
+; RV32-NEXT: sltu a1, s5, s4
; RV32-NEXT: add a1, a1, s11
; RV32-NEXT: add a0, a0, a1
-; RV32-NEXT: add s1, s0, s1
-; RV32-NEXT: sltu a1, s1, s0
-; RV32-NEXT: add a1, a1, ra
-; RV32-NEXT: add a0, a0, a1
; RV32-NEXT: vmv.x.s a1, v24
-; RV32-NEXT: add a2, s1, a2
-; RV32-NEXT: sltu a3, a2, s1
+; RV32-NEXT: add a2, s5, a2
+; RV32-NEXT: sltu a3, a2, s5
; RV32-NEXT: add a1, a3, a1
; RV32-NEXT: vmv.x.s a3, v16
; RV32-NEXT: add a0, a0, a1
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 3711f014e06478b..6c0288dd9b5d4ae 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
@@ -2431,7 +2431,7 @@ define <8 x i32> @mgather_v8i32(<8 x ptr> %ptrs, <8 x i1> %m, <8 x i32> %passthr
; RV64ZVE32F-NEXT: .LBB34_10: # %cond.load1
; RV64ZVE32F-NEXT: ld a2, 8(a0)
; RV64ZVE32F-NEXT: lw a2, 0(a2)
-; RV64ZVE32F-NEXT: vsetivli zero, 2, e32, m2, tu, ma
...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/66671
More information about the llvm-commits
mailing list