[rpms/kquickimageeditor] rawhide: Fix build failure on arm64

[rpms/kquickimageeditor] rawhide: Fix build failure on arm64

[Steve Cossette]

A new commit has been pushed.

Repo   : rpms/kquickimageeditor
Branch : rawhide
Commit : 5460730e5b7e56805fd1f03168139a69d5d8cdcf
Author : Steve Cossette <farchord@gmail.com>
Date   : 2026-09-02T07:00:30-04:00
Stats  : +110/-2 in 2 file(s)
URL    : https://src.fedoraproject.org/rpms/kquickimageeditor/c/5460730e5b7e56805fd1f03168139a69d5d8cdcf?branch=rawhide

Log:
Fix build failure on arm64

---
diff --git a/68.patch b/68.patch
new file mode 100644
index 0000000..7e0bfdf
--- /dev/null
+++ b/68.patch
@@ -0,0 +1,99 @@
+From ac7568044defc89c494b0d6edb60073cb0ad8b25 Mon Sep 17 00:00:00 2001
+From: Noah Davis <noahadvs@gmail.com>
+Date: Tue, 1 Sep 2026 12:45:18 -0400
+Subject: [PATCH] Fix build on ARM with SVE
+
+Can't put SIMD vectors in a struct/class or any container type on ARM with SVE since the SIMD vector sizes are only known at runtime.
+---
+ src/annotations/stackblur.cpp | 29 ++++++++++++++++-------------
+ 1 file changed, 16 insertions(+), 13 deletions(-)
+
+diff --git a/src/annotations/stackblur.cpp b/src/annotations/stackblur.cpp
+index ae4d750..af56c62 100644
+--- a/src/annotations/stackblur.cpp
++++ b/src/annotations/stackblur.cpp
+@@ -350,7 +350,6 @@ HWY_ATTR HWY_FLATTEN void blurRowWorker(const BlurTask &task)
+     // Vectors are what actually hold data, but they don't have any members.
+     // They don't even have a size until you zero them or assign something.
+     // You have to use Highway's APIs to do anything with them.
+-    using MVec = hn::Vec<MTag>; // math vector type
+ 
+     // Tags don't have data, Highway just uses them to pass type info.
+     MTag mtag;
+@@ -374,19 +373,22 @@ HWY_ATTR HWY_FLATTEN void blurRowWorker(const BlurTask &task)
+     // do this to multiply it with another vector.
+     const auto vMultiplier = hn::Set(mtag, radiusMultiplier<M_t>(radius));
+ 
++    // Can't put SIMD vectors in a struct/class or any container type on ARM
++    // with SVE since the SIMD vector sizes are only known at runtime.
++
+     // Weights should be set like {1, 2, 3, …, radius+1, …, 3, 2, 1}
+-    hwy::AlignedVector<MVec> weights(kernelSize);
++    hwy::AlignedVector<M_t> weights(kernelSize);
+     // set left side of kernel weights
+     for (Px i = 0; i < leftEndMidStart; ++i) {
+-        weights[i] = hn::Set(mtag, i + 1);
++        weights[i] = i + 1;
+     }
+     // set middle and right side of kernel weights
+     for (Px i = leftEndMidStart; i < kernelSize; ++i) {
+-        weights[i] = hn::Set(mtag, kernelSize - i);
++        weights[i] = kernelSize - i;
+     }
+ 
+-    // A queue of pixel data used by the kernel
+-    hwy::AlignedVector<MVec> stack(kernelSize);
++    // A queue of channel data used by the kernel
++    hwy::AlignedVector<M_t> stack(kernelSize * channels);
+ 
+     for (Px y = task.startRow; y < task.endRow; ++y) {
+         // Inform the compiler if the pointer is aligned and restricted
+@@ -408,18 +410,18 @@ HWY_ATTR HWY_FLATTEN void blurRowWorker(const BlurTask &task)
+ 
+         // fill up left side of kernel
+         for (Px i = 0; i < leftEndMidStart; ++i) {
+-            stack[i] = vFirst;
++            std::copy_n(&rowData[0], channels, &stack[i * channels]);
+             // MulAdd multiplies the first two args, then adds the last.
+             // Sometimes it's more optimal than Add(Mul(v0, v1), v2).
+-            stackSum = hn::MulAdd(vFirst, weights[i], stackSum);
++            stackSum = hn::MulAdd(vFirst, hn::Set(mtag, weights[i]), stackSum);
+             sumOut = hn::Add(sumOut, vFirst);
+         }
+         // fill up middle and right side of kernel
+         for (Px i = leftEndMidStart; i < kernelSize; ++i) {
+             const Px nextX = std::min(i - radius, lastX); // starts at 1
+             const auto vNext = loadPtrToVec<alignment>(ctag, itag, mtag, &rowData[nextX * channels]);
+-            stack[i] = vNext;
+-            stackSum = hn::MulAdd(vNext, weights[i], stackSum);
++            std::copy_n(&rowData[nextX * channels], channels, &stack[i * channels]);
++            stackSum = hn::MulAdd(vNext, hn::Set(mtag, weights[i]), stackSum);
+             sumIn = hn::Add(sumIn, vNext);
+         }
+ 
+@@ -441,7 +443,7 @@ HWY_ATTR HWY_FLATTEN void blurRowWorker(const BlurTask &task)
+                 // Outgoing data kernel index.
+                 // The modulo lets us loop through the kernel.
+                 const Px outKI = (kernelMidX - radius + kernelSize) % kernelSize;
+-                sumOut = hn::Sub(sumOut, stack[outKI]);
++                sumOut = hn::Sub(sumOut, hn::Load(mtag, &stack[outKI * channels]));
+                 // Next middle of kernel
+                 if constexpr (Section == LeftEdge) {
+                     kernelMidX = std::min(kernelMidX + 1, lastX);
+@@ -449,10 +451,11 @@ HWY_ATTR HWY_FLATTEN void blurRowWorker(const BlurTask &task)
+                     ++kernelMidX;
+                 }
+                 const auto vNextKMid = loadPtrToVec<alignment>(ctag, itag, mtag, &row[kernelMidX * channels]);
+-                stack[outKI] = vNextKMid; // outgoing is replaced by next
++                // outgoing is replaced by next
++                std::copy_n(&row[kernelMidX * channels], channels, &stack[outKI * channels]);
+                 sumIn = hn::Add(sumIn, vNextKMid);
+                 stackSum = hn::Add(stackSum, sumIn);
+-                const auto vCurrentKMid = stack[kernelMidX % kernelSize];
++                const auto vCurrentKMid = hn::Load(mtag, &stack[(kernelMidX % kernelSize) * channels]);
+                 sumOut = hn::Add(sumOut, vCurrentKMid);
+                 sumIn = hn::Sub(sumIn, vCurrentKMid);
+             }
+-- 
+GitLab
+

diff --git a/kquickimageeditor.spec b/kquickimageeditor.spec
index e0f5e32..3e5dfde 100644
--- a/kquickimageeditor.spec
+++ b/kquickimageeditor.spec
@@ -1,11 +1,17 @@
 Name:    kquickimageeditor
 Version: 0.7.0
-Release: 1%{?dist}
+Release: 2%{?dist}
 Summary: QtQuick components providing basic image editing capabilities
 License: BSD-2-Clause AND CC0-1.0 AND LGPL-2.0-or-later AND LGPL-2.1-only AND LGPL-2.1-or-later AND LGPL-3.0-only
 URL:     https://invent.kde.org/libraries/%{name}
 Source0: https://download.kde.org/stable/%{name}/%{name}-%{version}.tar.xz
 
+# UPSTREAM
+
+# Fix build on ARM with SVE
+# https://invent.kde.org/libraries/kquickimageeditor/-/merge_requests/68
+Patch0:  68.patch
+
 BuildRequires: extra-cmake-modules
 
 BuildRequires: kf6-rpm-macros
@@ -36,7 +42,7 @@ The %{name}-qt6-devel package contains cmake and mkspecs for developing
 applications that use %{name}-qt6.
 
 %prep
-%autosetup -n %{name}-%{version}
+%autosetup -n %{name}-%{version} -p1
 
 %build
 %cmake_kf6
@@ -58,6 +64,9 @@ applications that use %{name}-qt6.
 %{_kf6_archdatadir}/mkspecs/modules/qt_KQuickImageEditor.pri
 
 %changelog
+* Wed Sep 02 2026 Steve Cossette <farchord@gmail.com> - 0.7.0-2
+- Add fix to build on aarch64
+
 * Tue Sep 01 2026 Steve Cossette <farchord@gmail.com> - 0.7.0-1
 - 0.7.0