[LoongArch] Move widenShuffleMask before lowerVECTOR_SHUFFLE_XVPERMI to improve code quality

[ylzsx]

No description provided.

[llvmbot]

@llvm/pr-subscribers-backend-loongarch

Author: Zhaoxin Yang (ylzsx)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/164219.diff

2 Files Affected:

• (modified) llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp (+6)
• (modified) llvm/test/CodeGen/LoongArch/lasx/shuffle-as-permute-and-shuffle.ll (+3-9)

diff --git a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
index f7deeafc9ccfc..21e8179a304ef 100644
--- a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
+++ b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
@@ -2103,6 +2103,12 @@ static SDValue
 lowerVECTOR_SHUFFLE_XVPERMI(const SDLoc &DL, ArrayRef<int> Mask, MVT VT,
                             SDValue V1, SelectionDAG &DAG,
                             const LoongArchSubtarget &Subtarget) {
+  // Before selecting XVPERMI_D, always try to widen vectors to gain more
+  // optimization opportunities.
+  if (SDValue NewShuffle = widenShuffleMask(
+          DL, Mask, VT, V1, DAG.getUNDEF(V1.getValueType()), DAG))
+    return NewShuffle;
+
   // Only consider XVPERMI_D.
   if (Mask.size() != 4 || (VT != MVT::v4i64 && VT != MVT::v4f64))
     return SDValue();
diff --git a/llvm/test/CodeGen/LoongArch/lasx/shuffle-as-permute-and-shuffle.ll b/llvm/test/CodeGen/LoongArch/lasx/shuffle-as-permute-and-shuffle.ll
index 245f76472b844..7149cdb99063a 100644
--- a/llvm/test/CodeGen/LoongArch/lasx/shuffle-as-permute-and-shuffle.ll
+++ b/llvm/test/CodeGen/LoongArch/lasx/shuffle-as-permute-and-shuffle.ll
@@ -32,9 +32,7 @@ define <16 x i16> @shuffle_v16i16(<16 x i16> %a) {
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    pcalau12i $a0, %pc_hi20(.LCPI2_0)
 ; CHECK-NEXT:    xvld $xr1, $a0, %pc_lo12(.LCPI2_0)
-; CHECK-NEXT:    xvpermi.d $xr2, $xr0, 78
-; CHECK-NEXT:    xvshuf.w $xr1, $xr2, $xr0
-; CHECK-NEXT:    xvori.b $xr0, $xr1, 0
+; CHECK-NEXT:    xvperm.w $xr0, $xr0, $xr1
 ; CHECK-NEXT:    ret
   %shuffle = shufflevector <16 x i16> %a, <16 x i16> poison, <16 x i32> <i32 8, i32 9, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   ret <16 x i16> %shuffle
@@ -55,9 +53,7 @@ define <16 x i16> @shuffle_v16i16_same_lane(<16 x i16> %a) {
 define <8 x i32> @shuffle_v8i32(<8 x i32> %a) {
 ; CHECK-LABEL: shuffle_v8i32:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    pcalau12i $a0, %pc_hi20(.LCPI4_0)
-; CHECK-NEXT:    xvld $xr1, $a0, %pc_lo12(.LCPI4_0)
-; CHECK-NEXT:    xvperm.w $xr0, $xr0, $xr1
+; CHECK-NEXT:    xvpermi.d $xr0, $xr0, 226
 ; CHECK-NEXT:    ret
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> poison, <8 x i32> <i32 4, i32 5, i32 0, i32 1, i32 4, i32 5, i32 6, i32 7>
   ret <8 x i32> %shuffle
@@ -93,9 +89,7 @@ define <4 x i64> @shuffle_v4i64_same_lane(<4 x i64> %a) {
 define <8 x float> @shuffle_v8f32(<8 x float> %a) {
 ; CHECK-LABEL: shuffle_v8f32:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    pcalau12i $a0, %pc_hi20(.LCPI8_0)
-; CHECK-NEXT:    xvld $xr1, $a0, %pc_lo12(.LCPI8_0)
-; CHECK-NEXT:    xvperm.w $xr0, $xr0, $xr1
+; CHECK-NEXT:    xvpermi.d $xr0, $xr0, 226
 ; CHECK-NEXT:    ret
   %shuffle = shufflevector <8 x float> %a, <8 x float> poison, <8 x i32> <i32 4, i32 5, i32 0, i32 1, i32 4, i32 5, i32 6, i32 7>
   ret <8 x float> %shuffle

[zhaoqi5]

This not only affects XVPERMI_D, but also all later branches — consider moving it out and placing it just before calling lowerVECTOR_SHUFFLE_XVPERMI?

[ylzsx]

Yeah, you are right. My comment was not precise enough and I will modify it. However I'm not sure if there could be scenarios where it worsens, although running the llvm-test-suite did not reveal any issues.

[zhaoqi5]

LGTM