[RISCV] Pass undef VECTOR_SHUFFLE indices on to BUILD_VECTOR

Often when lowering vector shuffles, we split the shuffle into two
LHS/RHS shuffles which are then blended together. To do so we split the
original indices into two, indexed into each respective vector. These
two index vectors are then separately lowered as BUILD_VECTORs.

This patch forwards on any undef indices to the BUILD_VECTOR, rather
than having the VECTOR_SHUFFLE lowering decide on an optimal concrete
index. The motiviation for ths change is so that we don't duplicate
optimization logic between the two lowering methods and let BUILD_VECTOR
do what it does best.

Propagating undef in this way allows us, for example, to generate
`vid.v` to produce the LHS indices of commonly-used interleave-type
shuffles. I have designs on further optimizing interleave-type and other
common shuffle patterns in the near future.

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D104789

diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 5a7fd47..ab0d018 100644 (file)
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -1826,14 +1826,13 @@ static SDValue lowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG,
     bool SelectMaskVal = (MaskIndex < (int)NumElts) ^ InvertMask;
     MaskVals.push_back(DAG.getConstant(SelectMaskVal, DL, XLenVT));
     if (!IsSelect) {
-      bool IsLHS = MaskIndex < (int)NumElts;
-      // For "undef" elements of -1, shuffle in element 0 instead.
-      GatherIndicesLHS.push_back(
-          DAG.getConstant(IsLHS ? std::max(MaskIndex, 0) : 0, DL, XLenVT));
-      // TODO: If we're masking out unused elements anyway, it might produce
-      // better code if we use the most-common element index instead of 0.
+      bool IsLHSOrUndefIndex = MaskIndex < (int)NumElts;
+      GatherIndicesLHS.push_back(IsLHSOrUndefIndex && MaskIndex >= 0
+                                     ? DAG.getConstant(MaskIndex, DL, XLenVT)
+                                     : DAG.getUNDEF(XLenVT));
       GatherIndicesRHS.push_back(
-          DAG.getConstant(IsLHS ? 0 : MaskIndex - NumElts, DL, XLenVT));
+          IsLHSOrUndefIndex ? DAG.getUNDEF(XLenVT)
+                            : DAG.getConstant(MaskIndex - NumElts, DL, XLenVT));
     }
   }
 

diff --git a/llvm/test/CodeGen/RISCV/rvv/common-shuffle-patterns.ll b/llvm/test/CodeGen/RISCV/rvv/common-shuffle-patterns.ll
new file mode 100644 (file)
index 0000000..7692f29
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/rvv/common-shuffle-patterns.ll
@@ -0,0 +1,49 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv64 -mattr=+experimental-v -verify-machineinstrs -riscv-v-vector-bits-min=128 < %s | FileCheck %s
+
+target datalayout = "e-m:e-p:64:64-i64:64-i128:128-n64-S128"
+target triple = "riscv64-unknown-unknown-elf"
+
+define dso_local <16 x i16> @interleave(<8 x i16> %v0, <8 x i16> %v1) {
+; CHECK-LABEL: interleave:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    # kill: def $v8 killed $v8 killed $v8m2 def $v8m2
+; CHECK-NEXT:    vsetivli zero, 16, e16, m2, ta, mu
+; CHECK-NEXT:    vmv.v.i v26, 0
+; CHECK-NEXT:    vsetivli zero, 8, e16, m2, tu, mu
+; CHECK-NEXT:    vmv2r.v v28, v26
+; CHECK-NEXT:    vslideup.vi v28, v8, 0
+; CHECK-NEXT:    vsetivli zero, 8, e16, m1, ta, mu
+; CHECK-NEXT:    vmv.v.i v30, 0
+; CHECK-NEXT:    vsetivli zero, 16, e16, m2, tu, mu
+; CHECK-NEXT:    vslideup.vi v28, v30, 8
+; CHECK-NEXT:    lui a0, %hi(.LCPI0_0)
+; CHECK-NEXT:    addi a0, a0, %lo(.LCPI0_0)
+; CHECK-NEXT:    vsetvli zero, zero, e16, m2, ta, mu
+; CHECK-NEXT:    vle16.v v10, (a0)
+; CHECK-NEXT:    vmv1r.v v8, v9
+; CHECK-NEXT:    vrgather.vv v12, v28, v10
+; CHECK-NEXT:    vsetivli zero, 8, e16, m2, tu, mu
+; CHECK-NEXT:    vslideup.vi v26, v8, 0
+; CHECK-NEXT:    vsetivli zero, 16, e16, m2, tu, mu
+; CHECK-NEXT:    vslideup.vi v26, v30, 8
+; CHECK-NEXT:    vsetvli zero, zero, e16, m2, ta, mu
+; CHECK-NEXT:    vid.v v28
+; CHECK-NEXT:    vrgather.vv v8, v12, v28
+; CHECK-NEXT:    lui a0, 11
+; CHECK-NEXT:    addiw a0, a0, -1366
+; CHECK-NEXT:    vsetivli zero, 1, e16, mf4, ta, mu
+; CHECK-NEXT:    vmv.s.x v0, a0
+; CHECK-NEXT:    lui a0, %hi(.LCPI0_1)
+; CHECK-NEXT:    addi a0, a0, %lo(.LCPI0_1)
+; CHECK-NEXT:    vsetivli zero, 16, e16, m2, ta, mu
+; CHECK-NEXT:    vle16.v v28, (a0)
+; CHECK-NEXT:    vsetvli zero, zero, e16, m2, tu, mu
+; CHECK-NEXT:    vrgather.vv v8, v26, v28, v0.t
+; CHECK-NEXT:    ret
+entry:
+  %v2 = shufflevector <8 x i16> %v0, <8 x i16> poison, <16 x i32> <i32 0, i32 undef, i32 1, i32 undef, i32 2, i32 undef, i32 3, i32 undef, i32 4, i32 undef, i32 5, i32 undef, i32 6, i32 undef, i32 7, i32 undef>
+  %v3 = shufflevector <8 x i16> %v1, <8 x i16> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  %v4 = shufflevector <16 x i16> %v2, <16 x i16> %v3, <16 x i32> <i32 0, i32 16, i32 2, i32 17, i32 4, i32 18, i32 6, i32 19, i32 8, i32 20, i32 10, i32 21, i32 12, i32 22, i32 14, i32 23>
+  ret <16 x i16> %v4
+}

diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fp-buildvec.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fp-buildvec.ll
index 3b69eab..08a3d41 100644 (file)
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fp-buildvec.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fp-buildvec.ll
@@ -63,28 +63,22 @@ define <4 x float> @hang_when_merging_stores_after_legalization(<8 x float> %x,
 ; LMULMAX2-NEXT:    addi a0, zero, 2
 ; LMULMAX2-NEXT:    vsetivli zero, 1, e8, mf8, ta, mu
 ; LMULMAX2-NEXT:    vmv.s.x v0, a0
-; LMULMAX2-NEXT:    lui a0, %hi(.LCPI1_0)
-; LMULMAX2-NEXT:    addi a0, a0, %lo(.LCPI1_0)
 ; LMULMAX2-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
-; LMULMAX2-NEXT:    vle32.v v27, (a0)
+; LMULMAX2-NEXT:    vmv.v.i v27, 3
 ; LMULMAX2-NEXT:    vsetvli zero, zero, e32, m1, tu, mu
 ; LMULMAX2-NEXT:    vrgather.vv v26, v9, v27, v0.t
 ; LMULMAX2-NEXT:    vsetvli zero, zero, e32, m1, ta, mu
-; LMULMAX2-NEXT:    vrgather.vv v27, v10, v25
+; LMULMAX2-NEXT:    vrgather.vv v28, v10, v25
 ; LMULMAX2-NEXT:    addi a0, zero, 8
 ; LMULMAX2-NEXT:    vsetivli zero, 1, e8, mf8, ta, mu
 ; LMULMAX2-NEXT:    vmv.s.x v0, a0
-; LMULMAX2-NEXT:    lui a0, %hi(.LCPI1_1)
-; LMULMAX2-NEXT:    addi a0, a0, %lo(.LCPI1_1)
-; LMULMAX2-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
-; LMULMAX2-NEXT:    vle32.v v25, (a0)
-; LMULMAX2-NEXT:    vsetvli zero, zero, e32, m1, tu, mu
-; LMULMAX2-NEXT:    vrgather.vv v27, v11, v25, v0.t
+; LMULMAX2-NEXT:    vsetivli zero, 4, e32, m1, tu, mu
+; LMULMAX2-NEXT:    vrgather.vv v28, v11, v27, v0.t
 ; LMULMAX2-NEXT:    addi a0, zero, 3
 ; LMULMAX2-NEXT:    vsetivli zero, 1, e8, mf8, ta, mu
 ; LMULMAX2-NEXT:    vmv.s.x v0, a0
 ; LMULMAX2-NEXT:    vsetivli zero, 4, e32, m1, ta, mu
-; LMULMAX2-NEXT:    vmerge.vvm v8, v27, v26, v0
+; LMULMAX2-NEXT:    vmerge.vvm v8, v28, v26, v0
 ; LMULMAX2-NEXT:    ret
   %z = shufflevector <8 x float> %x, <8 x float> %y, <4 x i32> <i32 0, i32 7, i32 8, i32 15>
   ret <4 x float> %z

diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fp-shuffles.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fp-shuffles.ll
index b7be584..554ab97 100644 (file)
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fp-shuffles.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fp-shuffles.ll
@@ -133,43 +133,35 @@ define <4 x double> @vrgather_permute_shuffle_uv_v4f64(<4 x double> %x) {
 define <4 x double> @vrgather_shuffle_vv_v4f64(<4 x double> %x, <4 x double> %y) {
 ; RV32-LABEL: vrgather_shuffle_vv_v4f64:
 ; RV32:       # %bb.0:
-; RV32-NEXT:    addi a0, zero, 1
-; RV32-NEXT:    vsetivli zero, 4, e16, mf2, ta, mu
-; RV32-NEXT:    vmv.s.x v25, a0
-; RV32-NEXT:    vmv.v.i v28, 0
-; RV32-NEXT:    vsetvli zero, zero, e16, mf2, tu, mu
-; RV32-NEXT:    vslideup.vi v28, v25, 3
 ; RV32-NEXT:    lui a0, %hi(.LCPI6_0)
 ; RV32-NEXT:    addi a0, a0, %lo(.LCPI6_0)
-; RV32-NEXT:    vsetvli zero, zero, e16, mf2, ta, mu
+; RV32-NEXT:    vsetivli zero, 4, e16, mf2, ta, mu
 ; RV32-NEXT:    vle16.v v25, (a0)
 ; RV32-NEXT:    vsetvli zero, zero, e64, m2, ta, mu
 ; RV32-NEXT:    vrgatherei16.vv v26, v8, v25
 ; RV32-NEXT:    addi a0, zero, 8
 ; RV32-NEXT:    vsetivli zero, 1, e8, mf8, ta, mu
 ; RV32-NEXT:    vmv.s.x v0, a0
-; RV32-NEXT:    vsetivli zero, 4, e64, m2, tu, mu
-; RV32-NEXT:    vrgatherei16.vv v26, v10, v28, v0.t
+; RV32-NEXT:    vsetivli zero, 4, e16, mf2, ta, mu
+; RV32-NEXT:    vmv.v.i v25, 1
+; RV32-NEXT:    vsetvli zero, zero, e64, m2, tu, mu
+; RV32-NEXT:    vrgatherei16.vv v26, v10, v25, v0.t
 ; RV32-NEXT:    vmv2r.v v8, v26
 ; RV32-NEXT:    ret
 ;
 ; RV64-LABEL: vrgather_shuffle_vv_v4f64:
 ; RV64:       # %bb.0:
-; RV64-NEXT:    addi a0, zero, 1
-; RV64-NEXT:    vsetivli zero, 4, e64, m2, ta, mu
-; RV64-NEXT:    vmv.s.x v26, a0
-; RV64-NEXT:    vmv.v.i v28, 0
-; RV64-NEXT:    vsetvli zero, zero, e64, m2, tu, mu
-; RV64-NEXT:    vslideup.vi v28, v26, 3
 ; RV64-NEXT:    lui a0, %hi(.LCPI6_0)
 ; RV64-NEXT:    addi a0, a0, %lo(.LCPI6_0)
-; RV64-NEXT:    vsetvli zero, zero, e64, m2, ta, mu
-; RV64-NEXT:    vle64.v v30, (a0)
-; RV64-NEXT:    vrgather.vv v26, v8, v30
+; RV64-NEXT:    vsetivli zero, 4, e64, m2, ta, mu
+; RV64-NEXT:    vle64.v v28, (a0)
+; RV64-NEXT:    vrgather.vv v26, v8, v28
 ; RV64-NEXT:    addi a0, zero, 8
 ; RV64-NEXT:    vsetivli zero, 1, e8, mf8, ta, mu
 ; RV64-NEXT:    vmv.s.x v0, a0
-; RV64-NEXT:    vsetivli zero, 4, e64, m2, tu, mu
+; RV64-NEXT:    vsetivli zero, 4, e64, m2, ta, mu
+; RV64-NEXT:    vmv.v.i v28, 1
+; RV64-NEXT:    vsetvli zero, zero, e64, m2, tu, mu
 ; RV64-NEXT:    vrgather.vv v26, v10, v28, v0.t
 ; RV64-NEXT:    vmv2r.v v8, v26
 ; RV64-NEXT:    ret
@@ -185,28 +177,31 @@ define <4 x double> @vrgather_shuffle_xv_v4f64(<4 x double> %x) {
 ; RV32-NEXT:    vmv.s.x v0, a0
 ; RV32-NEXT:    lui a0, %hi(.LCPI7_0)
 ; RV32-NEXT:    addi a0, a0, %lo(.LCPI7_0)
-; RV32-NEXT:    vsetivli zero, 4, e16, mf2, ta, mu
-; RV32-NEXT:    vle16.v v25, (a0)
-; RV32-NEXT:    lui a0, %hi(.LCPI7_1)
-; RV32-NEXT:    addi a0, a0, %lo(.LCPI7_1)
-; RV32-NEXT:    vsetvli zero, zero, e64, m2, ta, mu
+; RV32-NEXT:    vsetivli zero, 4, e64, m2, ta, mu
 ; RV32-NEXT:    vlse64.v v26, (a0), zero
-; RV32-NEXT:    vsetvli zero, zero, e64, m2, tu, mu
+; RV32-NEXT:    lui a0, 16
+; RV32-NEXT:    addi a0, a0, 2
+; RV32-NEXT:    vsetivli zero, 2, e32, mf2, ta, mu
+; RV32-NEXT:    vmv.v.x v25, a0
+; RV32-NEXT:    vsetivli zero, 4, e64, m2, tu, mu
 ; RV32-NEXT:    vrgatherei16.vv v26, v8, v25, v0.t
 ; RV32-NEXT:    vmv2r.v v8, v26
 ; RV32-NEXT:    ret
 ;
 ; RV64-LABEL: vrgather_shuffle_xv_v4f64:
 ; RV64:       # %bb.0:
+; RV64-NEXT:    addi a0, zero, 2
+; RV64-NEXT:    vsetivli zero, 4, e64, m2, ta, mu
+; RV64-NEXT:    vmv.s.x v26, a0
+; RV64-NEXT:    vmv.v.i v28, 1
+; RV64-NEXT:    vsetivli zero, 3, e64, m2, tu, mu
+; RV64-NEXT:    vslideup.vi v28, v26, 2
 ; RV64-NEXT:    addi a0, zero, 12
 ; RV64-NEXT:    vsetivli zero, 1, e8, mf8, ta, mu
 ; RV64-NEXT:    vmv.s.x v0, a0
 ; RV64-NEXT:    lui a0, %hi(.LCPI7_0)
 ; RV64-NEXT:    addi a0, a0, %lo(.LCPI7_0)
 ; RV64-NEXT:    vsetivli zero, 4, e64, m2, ta, mu
-; RV64-NEXT:    vle64.v v28, (a0)
-; RV64-NEXT:    lui a0, %hi(.LCPI7_1)
-; RV64-NEXT:    addi a0, a0, %lo(.LCPI7_1)
 ; RV64-NEXT:    vlse64.v v26, (a0), zero
 ; RV64-NEXT:    vsetvli zero, zero, e64, m2, tu, mu
 ; RV64-NEXT:    vrgather.vv v26, v8, v28, v0.t
@@ -220,30 +215,27 @@ define <4 x double> @vrgather_shuffle_vx_v4f64(<4 x double> %x) {
 ; RV32-LABEL: vrgather_shuffle_vx_v4f64:
 ; RV32:       # %bb.0:
 ; RV32-NEXT:    addi a0, zero, 3
-; RV32-NEXT:    vsetivli zero, 4, e16, mf2, ta, mu
-; RV32-NEXT:    vmv.s.x v25, a0
-; RV32-NEXT:    vmv.v.i v28, 0
-; RV32-NEXT:    vsetivli zero, 2, e16, mf2, tu, mu
-; RV32-NEXT:    vslideup.vi v28, v25, 1
 ; RV32-NEXT:    vsetivli zero, 1, e8, mf8, ta, mu
 ; RV32-NEXT:    vmv.s.x v0, a0
 ; RV32-NEXT:    lui a0, %hi(.LCPI8_0)
 ; RV32-NEXT:    addi a0, a0, %lo(.LCPI8_0)
 ; RV32-NEXT:    vsetivli zero, 4, e64, m2, ta, mu
 ; RV32-NEXT:    vlse64.v v26, (a0), zero
-; RV32-NEXT:    vsetvli zero, zero, e64, m2, tu, mu
-; RV32-NEXT:    vrgatherei16.vv v26, v8, v28, v0.t
+; RV32-NEXT:    lui a0, 48
+; RV32-NEXT:    vsetivli zero, 2, e32, mf2, ta, mu
+; RV32-NEXT:    vmv.v.x v25, a0
+; RV32-NEXT:    vsetivli zero, 4, e64, m2, tu, mu
+; RV32-NEXT:    vrgatherei16.vv v26, v8, v25, v0.t
 ; RV32-NEXT:    vmv2r.v v8, v26
 ; RV32-NEXT:    ret
 ;
 ; RV64-LABEL: vrgather_shuffle_vx_v4f64:
 ; RV64:       # %bb.0:
-; RV64-NEXT:    addi a0, zero, 3
 ; RV64-NEXT:    vsetivli zero, 4, e64, m2, ta, mu
-; RV64-NEXT:    vmv.s.x v26, a0
-; RV64-NEXT:    vmv.v.i v28, 0
-; RV64-NEXT:    vsetivli zero, 2, e64, m2, tu, mu
-; RV64-NEXT:    vslideup.vi v28, v26, 1
+; RV64-NEXT:    vmv.v.i v28, 3
+; RV64-NEXT:    vsetvli zero, zero, e64, m2, tu, mu
+; RV64-NEXT:    vmv.s.x v28, zero
+; RV64-NEXT:    addi a0, zero, 3
 ; RV64-NEXT:    vsetivli zero, 1, e8, mf8, ta, mu
 ; RV64-NEXT:    vmv.s.x v0, a0
 ; RV64-NEXT:    lui a0, %hi(.LCPI8_0)

diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-shuffles.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-shuffles.ll
index 43cf0a6..280dc01 100644 (file)
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-shuffles.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-shuffles.ll
@@ -85,21 +85,17 @@ define <4 x i16> @vrgather_permute_shuffle_uv_v4i16(<4 x i16> %x) {
 define <4 x i16> @vrgather_shuffle_vv_v4i16(<4 x i16> %x, <4 x i16> %y) {
 ; CHECK-LABEL: vrgather_shuffle_vv_v4i16:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    addi a0, zero, 1
-; CHECK-NEXT:    vsetivli zero, 4, e16, mf2, ta, mu
-; CHECK-NEXT:    vmv.s.x v25, a0
-; CHECK-NEXT:    vmv.v.i v26, 0
-; CHECK-NEXT:    vsetvli zero, zero, e16, mf2, tu, mu
-; CHECK-NEXT:    vslideup.vi v26, v25, 3
 ; CHECK-NEXT:    lui a0, %hi(.LCPI6_0)
 ; CHECK-NEXT:    addi a0, a0, %lo(.LCPI6_0)
-; CHECK-NEXT:    vsetvli zero, zero, e16, mf2, ta, mu
-; CHECK-NEXT:    vle16.v v27, (a0)
-; CHECK-NEXT:    vrgather.vv v25, v8, v27
+; CHECK-NEXT:    vsetivli zero, 4, e16, mf2, ta, mu
+; CHECK-NEXT:    vle16.v v26, (a0)
+; CHECK-NEXT:    vrgather.vv v25, v8, v26
 ; CHECK-NEXT:    addi a0, zero, 8
 ; CHECK-NEXT:    vsetivli zero, 1, e8, mf8, ta, mu
 ; CHECK-NEXT:    vmv.s.x v0, a0
-; CHECK-NEXT:    vsetivli zero, 4, e16, mf2, tu, mu
+; CHECK-NEXT:    vsetivli zero, 4, e16, mf2, ta, mu
+; CHECK-NEXT:    vmv.v.i v26, 1
+; CHECK-NEXT:    vsetvli zero, zero, e16, mf2, tu, mu
 ; CHECK-NEXT:    vrgather.vv v25, v9, v26, v0.t
 ; CHECK-NEXT:    vmv1r.v v8, v25
 ; CHECK-NEXT:    ret
@@ -108,20 +104,37 @@ define <4 x i16> @vrgather_shuffle_vv_v4i16(<4 x i16> %x, <4 x i16> %y) {
 }
 
 define <4 x i16> @vrgather_shuffle_xv_v4i16(<4 x i16> %x) {
-; CHECK-LABEL: vrgather_shuffle_xv_v4i16:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    addi a0, zero, 12
-; CHECK-NEXT:    vsetivli zero, 1, e8, mf8, ta, mu
-; CHECK-NEXT:    vmv.s.x v0, a0
-; CHECK-NEXT:    lui a0, %hi(.LCPI7_0)
-; CHECK-NEXT:    addi a0, a0, %lo(.LCPI7_0)
-; CHECK-NEXT:    vsetivli zero, 4, e16, mf2, ta, mu
-; CHECK-NEXT:    vle16.v v26, (a0)
-; CHECK-NEXT:    vmv.v.i v25, 5
-; CHECK-NEXT:    vsetvli zero, zero, e16, mf2, tu, mu
-; CHECK-NEXT:    vrgather.vv v25, v8, v26, v0.t
-; CHECK-NEXT:    vmv1r.v v8, v25
-; CHECK-NEXT:    ret
+; RV32-LABEL: vrgather_shuffle_xv_v4i16:
+; RV32:       # %bb.0:
+; RV32-NEXT:    addi a0, zero, 12
+; RV32-NEXT:    vsetivli zero, 1, e8, mf8, ta, mu
+; RV32-NEXT:    vmv.s.x v0, a0
+; RV32-NEXT:    lui a0, 16
+; RV32-NEXT:    addi a0, a0, 2
+; RV32-NEXT:    vsetivli zero, 2, e32, mf2, ta, mu
+; RV32-NEXT:    vmv.v.x v26, a0
+; RV32-NEXT:    vsetivli zero, 4, e16, mf2, ta, mu
+; RV32-NEXT:    vmv.v.i v25, 5
+; RV32-NEXT:    vsetvli zero, zero, e16, mf2, tu, mu
+; RV32-NEXT:    vrgather.vv v25, v8, v26, v0.t
+; RV32-NEXT:    vmv1r.v v8, v25
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: vrgather_shuffle_xv_v4i16:
+; RV64:       # %bb.0:
+; RV64-NEXT:    addi a0, zero, 12
+; RV64-NEXT:    vsetivli zero, 1, e8, mf8, ta, mu
+; RV64-NEXT:    vmv.s.x v0, a0
+; RV64-NEXT:    lui a0, 16
+; RV64-NEXT:    addiw a0, a0, 2
+; RV64-NEXT:    vsetivli zero, 2, e32, mf2, ta, mu
+; RV64-NEXT:    vmv.v.x v26, a0
+; RV64-NEXT:    vsetivli zero, 4, e16, mf2, ta, mu
+; RV64-NEXT:    vmv.v.i v25, 5
+; RV64-NEXT:    vsetvli zero, zero, e16, mf2, tu, mu
+; RV64-NEXT:    vrgather.vv v25, v8, v26, v0.t
+; RV64-NEXT:    vmv1r.v v8, v25
+; RV64-NEXT:    ret
   %s = shufflevector <4 x i16> <i16 5, i16 5, i16 5, i16 5>, <4 x i16> %x, <4 x i32> <i32 0, i32 3, i32 6, i32 5>
   ret <4 x i16> %s
 }
@@ -130,13 +143,11 @@ define <4 x i16> @vrgather_shuffle_vx_v4i16(<4 x i16> %x) {
 ; CHECK-LABEL: vrgather_shuffle_vx_v4i16:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    addi a0, zero, 3
-; CHECK-NEXT:    vsetivli zero, 4, e16, mf2, ta, mu
-; CHECK-NEXT:    vmv.s.x v25, a0
-; CHECK-NEXT:    vmv.v.i v26, 0
-; CHECK-NEXT:    vsetivli zero, 2, e16, mf2, tu, mu
-; CHECK-NEXT:    vslideup.vi v26, v25, 1
 ; CHECK-NEXT:    vsetivli zero, 1, e8, mf8, ta, mu
 ; CHECK-NEXT:    vmv.s.x v0, a0
+; CHECK-NEXT:    lui a0, 48
+; CHECK-NEXT:    vsetivli zero, 2, e32, mf2, ta, mu
+; CHECK-NEXT:    vmv.v.x v26, a0
 ; CHECK-NEXT:    vsetivli zero, 4, e16, mf2, ta, mu
 ; CHECK-NEXT:    vmv.v.i v25, 5
 ; CHECK-NEXT:    vsetvli zero, zero, e16, mf2, tu, mu
@@ -203,12 +214,7 @@ define <8 x i64> @vrgather_shuffle_vv_v8i64(<8 x i64> %x, <8 x i64> %y) {
 ; RV32-NEXT:    addi a0, zero, 5
 ; RV32-NEXT:    vsetivli zero, 8, e16, m1, ta, mu
 ; RV32-NEXT:    vmv.s.x v25, a0
-; RV32-NEXT:    addi a0, zero, 36
-; RV32-NEXT:    vsetivli zero, 1, e8, mf8, ta, mu
-; RV32-NEXT:    vmv.s.x v0, a0
-; RV32-NEXT:    vsetivli zero, 8, e16, m1, ta, mu
-; RV32-NEXT:    vmv.v.i v26, 0
-; RV32-NEXT:    vmerge.vim v26, v26, 2, v0
+; RV32-NEXT:    vmv.v.i v26, 2
 ; RV32-NEXT:    vsetvli zero, zero, e16, m1, tu, mu
 ; RV32-NEXT:    vslideup.vi v26, v25, 7
 ; RV32-NEXT:    lui a0, %hi(.LCPI11_0)
@@ -230,12 +236,7 @@ define <8 x i64> @vrgather_shuffle_vv_v8i64(<8 x i64> %x, <8 x i64> %y) {
 ; RV64-NEXT:    addi a0, zero, 5
 ; RV64-NEXT:    vsetivli zero, 8, e64, m4, ta, mu
 ; RV64-NEXT:    vmv.s.x v28, a0
-; RV64-NEXT:    addi a0, zero, 36
-; RV64-NEXT:    vsetivli zero, 1, e8, mf8, ta, mu
-; RV64-NEXT:    vmv.s.x v0, a0
-; RV64-NEXT:    vsetivli zero, 8, e64, m4, ta, mu
-; RV64-NEXT:    vmv.v.i v16, 0
-; RV64-NEXT:    vmerge.vim v16, v16, 2, v0
+; RV64-NEXT:    vmv.v.i v16, 2
 ; RV64-NEXT:    vsetvli zero, zero, e64, m4, tu, mu
 ; RV64-NEXT:    vslideup.vi v16, v28, 7
 ; RV64-NEXT:    lui a0, %hi(.LCPI11_0)
@@ -257,16 +258,6 @@ define <8 x i64> @vrgather_shuffle_vv_v8i64(<8 x i64> %x, <8 x i64> %y) {
 define <8 x i64> @vrgather_shuffle_xv_v8i64(<8 x i64> %x) {
 ; RV32-LABEL: vrgather_shuffle_xv_v8i64:
 ; RV32:       # %bb.0:
-; RV32-NEXT:    addi a0, zero, 6
-; RV32-NEXT:    vsetivli zero, 8, e16, m1, ta, mu
-; RV32-NEXT:    vmv.s.x v25, a0
-; RV32-NEXT:    addi a0, zero, 4
-; RV32-NEXT:    vmv.s.x v26, a0
-; RV32-NEXT:    vmv.v.i v27, 0
-; RV32-NEXT:    vsetivli zero, 6, e16, m1, tu, mu
-; RV32-NEXT:    vslideup.vi v27, v26, 5
-; RV32-NEXT:    vsetivli zero, 7, e16, m1, tu, mu
-; RV32-NEXT:    vslideup.vi v27, v25, 6
 ; RV32-NEXT:    lui a0, %hi(.LCPI12_0)
 ; RV32-NEXT:    addi a0, a0, %lo(.LCPI12_0)
 ; RV32-NEXT:    vsetivli zero, 8, e16, m1, ta, mu
@@ -277,30 +268,27 @@ define <8 x i64> @vrgather_shuffle_xv_v8i64(<8 x i64> %x) {
 ; RV32-NEXT:    addi a0, zero, 113
 ; RV32-NEXT:    vsetivli zero, 1, e8, mf8, ta, mu
 ; RV32-NEXT:    vmv.s.x v0, a0
-; RV32-NEXT:    vsetivli zero, 8, e64, m4, tu, mu
-; RV32-NEXT:    vrgatherei16.vv v28, v8, v27, v0.t
+; RV32-NEXT:    lui a0, %hi(.LCPI12_1)
+; RV32-NEXT:    addi a0, a0, %lo(.LCPI12_1)
+; RV32-NEXT:    vsetivli zero, 8, e16, m1, ta, mu
+; RV32-NEXT:    vle16.v v25, (a0)
+; RV32-NEXT:    vsetvli zero, zero, e64, m4, tu, mu
+; RV32-NEXT:    vrgatherei16.vv v28, v8, v25, v0.t
 ; RV32-NEXT:    vmv4r.v v8, v28
 ; RV32-NEXT:    ret
 ;
 ; RV64-LABEL: vrgather_shuffle_xv_v8i64:
 ; RV64:       # %bb.0:
-; RV64-NEXT:    addi a0, zero, 6
-; RV64-NEXT:    vsetivli zero, 8, e64, m4, ta, mu
-; RV64-NEXT:    vmv.s.x v28, a0
-; RV64-NEXT:    addi a0, zero, 4
-; RV64-NEXT:    vmv.s.x v12, a0
-; RV64-NEXT:    vmv.v.i v16, 0
-; RV64-NEXT:    vsetivli zero, 6, e64, m4, tu, mu
-; RV64-NEXT:    vslideup.vi v16, v12, 5
-; RV64-NEXT:    vsetivli zero, 7, e64, m4, tu, mu
-; RV64-NEXT:    vslideup.vi v16, v28, 6
 ; RV64-NEXT:    addi a0, zero, 113
 ; RV64-NEXT:    vsetivli zero, 1, e8, mf8, ta, mu
 ; RV64-NEXT:    vmv.s.x v0, a0
+; RV64-NEXT:    lui a0, %hi(.LCPI12_0)
+; RV64-NEXT:    addi a0, a0, %lo(.LCPI12_0)
 ; RV64-NEXT:    vsetivli zero, 8, e64, m4, ta, mu
+; RV64-NEXT:    vle64.v v12, (a0)
 ; RV64-NEXT:    vmv.v.i v28, -1
 ; RV64-NEXT:    vsetvli zero, zero, e64, m4, tu, mu
-; RV64-NEXT:    vrgather.vv v28, v8, v16, v0.t
+; RV64-NEXT:    vrgather.vv v28, v8, v12, v0.t
 ; RV64-NEXT:    vmv4r.v v8, v28
 ; RV64-NEXT:    ret
   %s = shufflevector <8 x i64> <i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1>, <8 x i64> %x, <8 x i32> <i32 8, i32 3, i32 6, i32 5, i32 8, i32 12, i32 14, i32 3>
@@ -355,11 +343,10 @@ define <4 x i8> @interleave_shuffles(<4 x i8> %x) {
 ; CHECK-NEXT:    vmv.x.s a0, v8
 ; CHECK-NEXT:    vsetivli zero, 4, e8, mf4, ta, mu
 ; CHECK-NEXT:    vrgather.vi v25, v8, 1
-; CHECK-NEXT:    addi a1, zero, 1
-; CHECK-NEXT:    vmv.s.x v26, a1
-; CHECK-NEXT:    vmv.v.i v27, 0
-; CHECK-NEXT:    vsetvli zero, zero, e8, mf4, tu, mu
-; CHECK-NEXT:    vslideup.vi v27, v26, 3
+; CHECK-NEXT:    vmv.s.x v26, zero
+; CHECK-NEXT:    vmv.v.i v27, 1
+; CHECK-NEXT:    vsetivli zero, 2, e8, mf4, tu, mu
+; CHECK-NEXT:    vslideup.vi v27, v26, 1
 ; CHECK-NEXT:    addi a1, zero, 10
 ; CHECK-NEXT:    vsetivli zero, 1, e8, mf8, ta, mu
 ; CHECK-NEXT:    vmv.s.x v0, a1

diff --git a/llvm/test/CodeGen/RISCV/rvv/interleave-crash.ll b/llvm/test/CodeGen/RISCV/rvv/interleave-crash.ll
index 8d0fe18..33fad2a 100644 (file)
--- a/llvm/test/CodeGen/RISCV/rvv/interleave-crash.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/interleave-crash.ll
@@ -28,10 +28,8 @@ define void @interleave256(<256 x i16>* %agg.result, <128 x i16>* %0, <128 x i16
 ; RV64-1024-NEXT:    vslideup.vi v28, v8, 0
 ; RV64-1024-NEXT:    vsetvli zero, a1, e16, m4, tu, mu
 ; RV64-1024-NEXT:    vslideup.vx v28, v12, a3
-; RV64-1024-NEXT:    lui a2, %hi(.LCPI0_1)
-; RV64-1024-NEXT:    addi a2, a2, %lo(.LCPI0_1)
 ; RV64-1024-NEXT:    vsetvli zero, zero, e16, m4, ta, mu
-; RV64-1024-NEXT:    vle16.v v12, (a2)
+; RV64-1024-NEXT:    vid.v v12
 ; RV64-1024-NEXT:    vrgather.vv v8, v24, v12
 ; RV64-1024-NEXT:    lui a2, 1026731
 ; RV64-1024-NEXT:    addiw a2, a2, -1365
@@ -50,8 +48,8 @@ define void @interleave256(<256 x i16>* %agg.result, <128 x i16>* %0, <128 x i16
 ; RV64-1024-NEXT:    vslideup.vi v0, v25, 2
 ; RV64-1024-NEXT:    vsetivli zero, 4, e64, m1, tu, mu
 ; RV64-1024-NEXT:    vslideup.vi v0, v25, 3
-; RV64-1024-NEXT:    lui a2, %hi(.LCPI0_2)
-; RV64-1024-NEXT:    addi a2, a2, %lo(.LCPI0_2)
+; RV64-1024-NEXT:    lui a2, %hi(.LCPI0_1)
+; RV64-1024-NEXT:    addi a2, a2, %lo(.LCPI0_1)
 ; RV64-1024-NEXT:    vsetvli zero, a1, e16, m4, ta, mu
 ; RV64-1024-NEXT:    vle16.v v12, (a2)
 ; RV64-1024-NEXT:    vsetvli zero, zero, e16, m4, tu, mu
@@ -85,10 +83,8 @@ define void @interleave256(<256 x i16>* %agg.result, <128 x i16>* %0, <128 x i16
 ; RV64-2048-NEXT:    vslideup.vi v26, v30, 0
 ; RV64-2048-NEXT:    vsetvli zero, a1, e16, m2, tu, mu
 ; RV64-2048-NEXT:    vslideup.vx v26, v28, a3
-; RV64-2048-NEXT:    lui a2, %hi(.LCPI0_1)
-; RV64-2048-NEXT:    addi a2, a2, %lo(.LCPI0_1)
 ; RV64-2048-NEXT:    vsetvli zero, zero, e16, m2, ta, mu
-; RV64-2048-NEXT:    vle16.v v28, (a2)
+; RV64-2048-NEXT:    vid.v v28
 ; RV64-2048-NEXT:    vrgather.vv v30, v12, v28
 ; RV64-2048-NEXT:    lui a2, 1026731
 ; RV64-2048-NEXT:    addiw a2, a2, -1365
@@ -107,8 +103,8 @@ define void @interleave256(<256 x i16>* %agg.result, <128 x i16>* %0, <128 x i16
 ; RV64-2048-NEXT:    vslideup.vi v0, v25, 2
 ; RV64-2048-NEXT:    vsetivli zero, 4, e64, m1, tu, mu
 ; RV64-2048-NEXT:    vslideup.vi v0, v25, 3
-; RV64-2048-NEXT:    lui a2, %hi(.LCPI0_2)
-; RV64-2048-NEXT:    addi a2, a2, %lo(.LCPI0_2)
+; RV64-2048-NEXT:    lui a2, %hi(.LCPI0_1)
+; RV64-2048-NEXT:    addi a2, a2, %lo(.LCPI0_1)
 ; RV64-2048-NEXT:    vsetvli zero, a1, e16, m2, ta, mu
 ; RV64-2048-NEXT:    vle16.v v28, (a2)
 ; RV64-2048-NEXT:    vsetvli zero, zero, e16, m2, tu, mu
@@ -177,10 +173,8 @@ define void @interleave512(<512 x i16>* %agg.result, <256 x i16>* %0, <256 x i16
 ; RV64-1024-NEXT:    addi a2, sp, 16
 ; RV64-1024-NEXT:    vl8re8.v v16, (a2) # Unknown-size Folded Reload
 ; RV64-1024-NEXT:    vslideup.vx v8, v16, a3
-; RV64-1024-NEXT:    lui a2, %hi(.LCPI1_1)
-; RV64-1024-NEXT:    addi a2, a2, %lo(.LCPI1_1)
 ; RV64-1024-NEXT:    vsetvli zero, zero, e16, m8, ta, mu
-; RV64-1024-NEXT:    vle16.v v24, (a2)
+; RV64-1024-NEXT:    vid.v v24
 ; RV64-1024-NEXT:    csrr a2, vlenb
 ; RV64-1024-NEXT:    slli a2, a2, 3
 ; RV64-1024-NEXT:    add a2, sp, a2
@@ -212,8 +206,8 @@ define void @interleave512(<512 x i16>* %agg.result, <256 x i16>* %0, <256 x i16
 ; RV64-1024-NEXT:    vslideup.vi v0, v25, 6
 ; RV64-1024-NEXT:    vsetivli zero, 8, e64, m1, tu, mu
 ; RV64-1024-NEXT:    vslideup.vi v0, v25, 7
-; RV64-1024-NEXT:    lui a2, %hi(.LCPI1_2)
-; RV64-1024-NEXT:    addi a2, a2, %lo(.LCPI1_2)
+; RV64-1024-NEXT:    lui a2, %hi(.LCPI1_1)
+; RV64-1024-NEXT:    addi a2, a2, %lo(.LCPI1_1)
 ; RV64-1024-NEXT:    vsetvli zero, a1, e16, m8, ta, mu
 ; RV64-1024-NEXT:    vle16.v v24, (a2)
 ; RV64-1024-NEXT:    vsetvli zero, zero, e16, m8, tu, mu
@@ -252,10 +246,8 @@ define void @interleave512(<512 x i16>* %agg.result, <256 x i16>* %0, <256 x i16
 ; RV64-2048-NEXT:    vslideup.vi v28, v8, 0
 ; RV64-2048-NEXT:    vsetvli zero, a1, e16, m4, tu, mu
 ; RV64-2048-NEXT:    vslideup.vx v28, v12, a3
-; RV64-2048-NEXT:    lui a2, %hi(.LCPI1_1)
-; RV64-2048-NEXT:    addi a2, a2, %lo(.LCPI1_1)
 ; RV64-2048-NEXT:    vsetvli zero, zero, e16, m4, ta, mu
-; RV64-2048-NEXT:    vle16.v v12, (a2)
+; RV64-2048-NEXT:    vid.v v12
 ; RV64-2048-NEXT:    vrgather.vv v8, v24, v12
 ; RV64-2048-NEXT:    lui a2, 1026731
 ; RV64-2048-NEXT:    addiw a2, a2, -1365
@@ -282,8 +274,8 @@ define void @interleave512(<512 x i16>* %agg.result, <256 x i16>* %0, <256 x i16
 ; RV64-2048-NEXT:    vslideup.vi v0, v25, 6
 ; RV64-2048-NEXT:    vsetivli zero, 8, e64, m1, tu, mu
 ; RV64-2048-NEXT:    vslideup.vi v0, v25, 7
-; RV64-2048-NEXT:    lui a2, %hi(.LCPI1_2)
-; RV64-2048-NEXT:    addi a2, a2, %lo(.LCPI1_2)
+; RV64-2048-NEXT:    lui a2, %hi(.LCPI1_1)
+; RV64-2048-NEXT:    addi a2, a2, %lo(.LCPI1_1)
 ; RV64-2048-NEXT:    vsetvli zero, a1, e16, m4, ta, mu
 ; RV64-2048-NEXT:    vle16.v v12, (a2)
 ; RV64-2048-NEXT:    vsetvli zero, zero, e16, m4, tu, mu