[llvm] [DAG] isKnownToBeAPowerOfTwo - add ISD::VECTOR_SHUFFLE handling (PR #185203)
Arjun Parmar via llvm-commits
llvm-commits at lists.llvm.org
Sun Mar 8 04:41:14 PDT 2026
https://github.com/akparmar004 updated https://github.com/llvm/llvm-project/pull/185203
>From 18161b6a584800516423a9df25cd6d9b1f53c411 Mon Sep 17 00:00:00 2001
From: akparmar004 <akparmar0404 at gmail.com>
Date: Sat, 7 Mar 2026 22:21:33 +0530
Subject: [PATCH] [DAG] isKnownToBeAPowerOfTwo - add ISD::VECTOR_SHUFFLE
handling
---
.../lib/CodeGen/SelectionDAG/SelectionDAG.cpp | 24 +++++++++++++
llvm/test/CodeGen/X86/known-pow2.ll | 34 +++++++++++++++++++
llvm/test/CodeGen/X86/known-powof2-shuffle.ll | 12 +++++++
3 files changed, 70 insertions(+)
create mode 100644 llvm/test/CodeGen/X86/known-powof2-shuffle.ll
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index a2318fc034fa2..4ec771d7fd41f 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -4784,6 +4784,30 @@ bool SelectionDAG::isKnownToBeAPowerOfTwo(SDValue Val,
// vscale(power-of-two) is a power-of-two
return isKnownToBeAPowerOfTwo(Val.getOperand(0), /*OrZero=*/false,
Depth + 1);
+
+ case ISD::VECTOR_SHUFFLE: {
+ assert(!Val.getValueType().isScalableVector());
+ // Demanded elements with undef shuffle mask elements are unknown
+ // - we cannot guarantee they are a power of two, so return false.
+ APInt DemandedLHS, DemandedRHS;
+ const ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(Val);
+ assert(NumElts == SVN->getMask().size() && "Unexpected vector size");
+ if (!getShuffleDemandedElts(NumElts, SVN->getMask(), DemandedElts,
+ DemandedLHS, DemandedRHS))
+ return false;
+
+ // All demanded elements from LHS must be known power of two.
+ if (!!DemandedLHS && !isKnownToBeAPowerOfTwo(Val.getOperand(0), DemandedLHS,
+ OrZero, Depth + 1))
+ return false;
+
+ // All demanded elements from RHS must be known power of two.
+ if (!!DemandedRHS && !isKnownToBeAPowerOfTwo(Val.getOperand(1), DemandedRHS,
+ OrZero, Depth + 1))
+ return false;
+
+ return true;
+ }
}
// More could be done here, though the above checks are enough
diff --git a/llvm/test/CodeGen/X86/known-pow2.ll b/llvm/test/CodeGen/X86/known-pow2.ll
index a9b28458e9518..2fccbb088fb98 100644
--- a/llvm/test/CodeGen/X86/known-pow2.ll
+++ b/llvm/test/CodeGen/X86/known-pow2.ll
@@ -1241,3 +1241,37 @@ define i32 @pow2_rotr_extract_vec(<4 x i32> %a0, <4 x i32> %rotamt, i32 %x, ptr
%res = urem i32 %x, %elt
ret i32 %res
}
+
+define <4 x i32> @pow2_shuffle_vec(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) {
+; CHECK-LABEL: pow2_shuffle_vec:
+; CHECK: # %bb.0:
+; CHECK-NEXT: pxor %xmm3, %xmm3
+; CHECK-NEXT: pxor %xmm4, %xmm4
+; CHECK-NEXT: pcmpgtd %xmm0, %xmm4
+; CHECK-NEXT: pcmpgtd %xmm1, %xmm3
+; CHECK-NEXT: movdqa %xmm4, %xmm0
+; CHECK-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; CHECK-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4
+; CHECK-NEXT: por %xmm0, %xmm4
+; CHECK-NEXT: movdqa %xmm3, %xmm0
+; CHECK-NEXT: pandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; CHECK-NEXT: pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
+; CHECK-NEXT: por %xmm0, %xmm3
+; CHECK-NEXT: pshufd {{.*#+}} xmm0 = xmm3[1,3,2,3]
+; CHECK-NEXT: pshufd {{.*#+}} xmm1 = xmm4[0,2,2,3]
+; CHECK-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; CHECK-NEXT: pcmpeqd %xmm0, %xmm0
+; CHECK-NEXT: paddd %xmm1, %xmm0
+; CHECK-NEXT: pand %xmm2, %xmm0
+; CHECK-NEXT: retq
+ %cmp0 = icmp sgt <4 x i32> zeroinitializer, %a0
+ %cmp1 = icmp sgt <4 x i32> zeroinitializer, %a1
+ %sel0 = select <4 x i1> %cmp0, <4 x i32> <i32 4, i32 4, i32 4, i32 4>,
+ <4 x i32> <i32 8, i32 8, i32 8, i32 8>
+ %sel1 = select <4 x i1> %cmp1, <4 x i32> <i32 2, i32 2, i32 2, i32 2>,
+ <4 x i32> <i32 16, i32 16, i32 16, i32 16>
+ %shuf = shufflevector <4 x i32> %sel0, <4 x i32> %sel1,
+ <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+ %res = urem <4 x i32> %a2, %shuf
+ ret <4 x i32> %res
+}
diff --git a/llvm/test/CodeGen/X86/known-powof2-shuffle.ll b/llvm/test/CodeGen/X86/known-powof2-shuffle.ll
new file mode 100644
index 0000000000000..fff81fecb879e
--- /dev/null
+++ b/llvm/test/CodeGen/X86/known-powof2-shuffle.ll
@@ -0,0 +1,12 @@
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -o - %s | FileCheck %s
+
+; Test that isKnownToBeAPowerOfTwo handles VECTOR_SHUFFLE
+
+; A shuffle of two power-of-2 vectors should be recognized as power-of-2
+define <4 x i32> @shuffle_pow2(<4 x i32> %a) {
+; CHECK-LABEL: shuffle_pow2:
+ %pow2 = and <4 x i32> %a, <i32 4, i32 4, i32 4, i32 4>
+ %splat = shufflevector <4 x i32> %pow2, <4 x i32> poison,
+ <4 x i32> <i32 0, i32 0, i32 0, i32 0>
+ ret <4 x i32> %splat
+}
More information about the llvm-commits
mailing list