https://github.com/llvmbot updated
https://github.com/llvm/llvm-project/pull/125590
>From 820c8c79c87532b44c62a53cc3865cda8a0846ee Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.top...@sifive.com>
Date: Mon, 3 Feb 2025 13:48:42 -0800
Subject: [PATCH] [RISCV] Check isFixedLengthVector before calling
getVectorNumElements in getSingleShuffleSrc. (#125455)
I have been unsuccessful at further reducing the test. The
failure requires a shuffle with 2 scalable->fixed extracts with
the same source. 0 is the only valid index for a scalable->fixed
extract so the 2 sources must be the same extract. Shuffles with
the same source are aggressively canonicalized to a unary shuffle.
So it requires the extracts to become identical through other
optimizations without the shuffle being canonicalized before it is
lowered.
Fixes #125306.
(cherry picked from commit 7c5100d36d8027dd205d6ec410a63c3930a1d9c1)
---
llvm/lib/Target/RISCV/RISCVISelLowering.cpp | 3 +-
llvm/test/CodeGen/RISCV/rvv/pr125306.ll | 118 ++++++++++++++++++++
2 files changed, 120 insertions(+), 1 deletion(-)
create mode 100644 llvm/test/CodeGen/RISCV/rvv/pr125306.ll
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 8b5ee3e67ce637c..585af30f8c8a169 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -4512,7 +4512,8 @@ static SDValue getSingleShuffleSrc(MVT VT, MVT
ContainerVT, SDValue V1,
// Src needs to have twice the number of elements.
unsigned NumElts = VT.getVectorNumElements();
- if (Src.getValueType().getVectorNumElements() != (NumElts * 2))
+ if (!Src.getValueType().isFixedLengthVector() ||
+ Src.getValueType().getVectorNumElements() != (NumElts * 2))
return SDValue();
// The extracts must extract the two halves of the source.
diff --git a/llvm/test/CodeGen/RISCV/rvv/pr125306.ll
b/llvm/test/CodeGen/RISCV/rvv/pr125306.ll
new file mode 100644
index 000000000000000..111f87de220dbfa
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/rvv/pr125306.ll
@@ -0,0 +1,118 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
UTC_ARGS: --version 5
+; RUN: llc < %s -mtriple=riscv64 -mattr=+m,+v | FileCheck %s
+
+; Test for an "Invalid size request on a scalable vector". Attempts to reduce
+; the test faurther were not successful. The failure requires a shuffle with 2
+; scalable->fixed extracts from the same vector. 0 is the only valid index for
a
+; scalable->fixed extract so the 2 extract must be the same. Shuffles with the
+; same source are aggressively canonicalized to a unary shuffle so it requires
+; the extracts to become identical through other optimizations without the
+; shuffle being canonicalized before it is lowered.
+
+define <2 x i32> @main(ptr %0) {
+; CHECK-LABEL: main:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: vsetivli zero, 16, e32, m4, ta, ma
+; CHECK-NEXT: vmv.v.i v8, 0
+; CHECK-NEXT: vse32.v v8, (zero)
+; CHECK-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; CHECK-NEXT: vmv.v.i v8, 0
+; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; CHECK-NEXT: vmv.v.i v10, 0
+; CHECK-NEXT: li a2, 64
+; CHECK-NEXT: sw zero, 80(zero)
+; CHECK-NEXT: lui a1, 7
+; CHECK-NEXT: lui a3, 1
+; CHECK-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; CHECK-NEXT: vid.v v11
+; CHECK-NEXT: li a4, 16
+; CHECK-NEXT: lui a5, 2
+; CHECK-NEXT: vsetivli zero, 4, e32, m1, ta, ma
+; CHECK-NEXT: vse32.v v10, (a2)
+; CHECK-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; CHECK-NEXT: vmv.v.i v10, 0
+; CHECK-NEXT: li a2, 24
+; CHECK-NEXT: sh zero, -392(a3)
+; CHECK-NEXT: sh zero, 534(a3)
+; CHECK-NEXT: sh zero, 1460(a3)
+; CHECK-NEXT: li a3, 32
+; CHECK-NEXT: vse32.v v10, (a2)
+; CHECK-NEXT: li a2, 40
+; CHECK-NEXT: vsetivli zero, 8, e32, m2, ta, ma
+; CHECK-NEXT: vse32.v v8, (a0)
+; CHECK-NEXT: sh zero, -1710(a5)
+; CHECK-NEXT: sh zero, -784(a5)
+; CHECK-NEXT: sh zero, 142(a5)
+; CHECK-NEXT: lw a5, -304(a1)
+; CHECK-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
+; CHECK-NEXT: vadd.vi v9, v11, -1
+; CHECK-NEXT: vse32.v v10, (a3)
+; CHECK-NEXT: sh zero, 0(a0)
+; CHECK-NEXT: lw a0, -188(a1)
+; CHECK-NEXT: vse32.v v10, (a2)
+; CHECK-NEXT: lw a2, -188(a1)
+; CHECK-NEXT: lw a3, 1244(a1)
+; CHECK-NEXT: vmv.v.x v8, a0
+; CHECK-NEXT: lw a0, 1244(a1)
+; CHECK-NEXT: lw a1, -304(a1)
+; CHECK-NEXT: vmv.v.x v10, a3
+; CHECK-NEXT: vmv.v.x v11, a5
+; CHECK-NEXT: vslide1down.vx v8, v8, zero
+; CHECK-NEXT: vslide1down.vx v10, v10, zero
+; CHECK-NEXT: vmin.vv v8, v10, v8
+; CHECK-NEXT: vmv.v.x v10, a0
+; CHECK-NEXT: vslide1down.vx v11, v11, zero
+; CHECK-NEXT: vmin.vx v10, v10, a2
+; CHECK-NEXT: vmin.vx v10, v10, a1
+; CHECK-NEXT: vmin.vv v11, v8, v11
+; CHECK-NEXT: vmv1r.v v8, v10
+; CHECK-NEXT: vand.vv v9, v11, v9
+; CHECK-NEXT: vslideup.vi v8, v10, 1
+; CHECK-NEXT: vse32.v v9, (a4)
+; CHECK-NEXT: sh zero, 0(zero)
+; CHECK-NEXT: ret
+entry:
+ store <16 x i32> zeroinitializer, ptr null, align 4
+ store <8 x i32> zeroinitializer, ptr %0, align 4
+ store <4 x i32> zeroinitializer, ptr getelementptr inbounds nuw (i8, ptr
null, i64 64), align 4
+ store i32 0, ptr getelementptr inbounds nuw (i8, ptr null, i64 80), align 4
+ %1 = load i32, ptr getelementptr inbounds nuw (i8, ptr null, i64 29916),
align 4
+ %broadcast.splatinsert53 = insertelement <4 x i32> zeroinitializer, i32 %1,
i64 0
+ %2 = load i32, ptr getelementptr inbounds nuw (i8, ptr null, i64 28484),
align 4
+ %broadcast.splatinsert55 = insertelement <4 x i32> zeroinitializer, i32 %2,
i64 0
+ %3 = call <4 x i32> @llvm.smin.v4i32(<4 x i32> %broadcast.splatinsert53, <4
x i32> %broadcast.splatinsert55)
+ %4 = load i32, ptr getelementptr inbounds nuw (i8, ptr null, i64 28368),
align 4
+ %broadcast.splatinsert57 = insertelement <4 x i32> zeroinitializer, i32 %4,
i64 0
+ %5 = call <4 x i32> @llvm.smin.v4i32(<4 x i32> %3, <4 x i32>
%broadcast.splatinsert57)
+ store i16 0, ptr getelementptr inbounds nuw (i8, ptr null, i64 3704), align 2
+ store i16 0, ptr getelementptr inbounds nuw (i8, ptr null, i64 4630), align 2
+ %6 = shufflevector <4 x i32> %5, <4 x i32> zeroinitializer, <2 x i32> <i32
0, i32 4>
+ store <2 x i32> %6, ptr getelementptr inbounds nuw (i8, ptr null, i64 16),
align 4
+ store i16 0, ptr getelementptr inbounds nuw (i8, ptr null, i64 5556), align 2
+ store i16 0, ptr getelementptr inbounds nuw (i8, ptr null, i64 6482), align 2
+ store <2 x i32> zeroinitializer, ptr getelementptr inbounds nuw (i8, ptr
null, i64 24), align 4
+ store i16 0, ptr getelementptr inbounds nuw (i8, ptr null, i64 7408), align 2
+ store i16 0, ptr getelementptr inbounds nuw (i8, ptr null, i64 8334), align 2
+ store <2 x i32> zeroinitializer, ptr getelementptr inbounds nuw (i8, ptr
null, i64 32), align 4
+ store i16 0, ptr %0, align 2
+ store <2 x i32> zeroinitializer, ptr getelementptr inbounds nuw (i8, ptr
null, i64 40), align 4
+ %7 = load i32, ptr getelementptr inbounds nuw (i8, ptr null, i64 29916),
align 4
+ %broadcast.splatinsert165 = insertelement <4 x i32> poison, i32 %7, i64 0
+ %8 = load i32, ptr getelementptr inbounds nuw (i8, ptr null, i64 28484),
align 4
+ %broadcast.splatinsert167 = insertelement <4 x i32> poison, i32 %8, i64 0
+ %9 = call <4 x i32> @llvm.smin.v4i32(<4 x i32> %broadcast.splatinsert165, <4
x i32> %broadcast.splatinsert167)
+ %10 = load i32, ptr getelementptr inbounds nuw (i8, ptr null, i64 28368),
align 4
+ %broadcast.splatinsert169 = insertelement <4 x i32> poison, i32 %10, i64 0
+ %11 = call <4 x i32> @llvm.smin.v4i32(<4 x i32> %9, <4 x i32>
%broadcast.splatinsert169)
+ store i16 0, ptr null, align 2
+ %12 = load i32, ptr getelementptr inbounds nuw (i8, ptr null, i64 29916),
align 4
+ %broadcast.splatinsert179 = insertelement <4 x i32> poison, i32 %12, i64 0
+ %13 = load i32, ptr getelementptr inbounds nuw (i8, ptr null, i64 28484),
align 4
+ %broadcast.splatinsert181 = insertelement <4 x i32> poison, i32 %13, i64 0
+ %14 = call <4 x i32> @llvm.smin.v4i32(<4 x i32> %broadcast.splatinsert179,
<4 x i32> %broadcast.splatinsert181)
+ %15 = load i32, ptr getelementptr inbounds nuw (i8, ptr null, i64 28368),
align 4
+ %broadcast.splatinsert183 = insertelement <4 x i32> poison, i32 %15, i64 0
+ %16 = call <4 x i32> @llvm.smin.v4i32(<4 x i32> %14, <4 x i32>
%broadcast.splatinsert183)
+ %17 = shufflevector <4 x i32> %11, <4 x i32> %16, <2 x i32> <i32 0, i32 4>
+ ret <2 x i32> %17
+}
_______________________________________________
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
