Update ASTC encoder to 5.3.0 (#1036)

5.3.0 is a minor maintenance release.

git subrepo commit (merge) external/astc-encoder

subrepo:
  subdir:   "external/astc-encoder"
  merged:   "1c5233bf0"
upstream:
  origin:   "https://github.com/ARM-software/astc-encoder.git"
  branch:   "5.2.0"
  commit:   "30aabb3f4"
git-subrepo:
  version:  "0.4.9"
  origin:   "https://github.com/MarkCallow/git-subrepo.git"
  commit:   "4f60dd7"

Author: MarkCallow

external/astc-encoder/.github/workflows/build_test.yaml

@@ -17,8 +17,11 @@ jobs:
         with:
           submodules: 'true'
 
+      - name: Update apt packages
+        run: sudo apt-get update
+
       - name: Install ImageMagick
-        run: sudo apt install imagemagick
+        run: sudo apt-get install imagemagick
 
       - name: Build release
         run: |
@@ -55,18 +58,16 @@ jobs:
           pip install numpy Pillow
 
       - name: Run system tests
-        # GitHub is using Cobalt-100, which has 128-bit SVE so we cannot test sve_256 here
+        # Disable SVE for now
         run: |
           python ./Test/astc_test_functional.py --encoder none
           python ./Test/astc_test_functional.py --encoder neon
-          python ./Test/astc_test_functional.py --encoder sve_128
           python ./Test/astc_test_image.py --encoder none --test-set Small
           python ./Test/astc_test_image.py --encoder neon --test-set Small
-          python ./Test/astc_test_image.py --encoder sve_128 --test-set Small
 
       - name: Run unit tests
-        # GitHub is using Cobalt-100, which has 128-bit SVE so we cannot test sve_256 here
-        run: ctest -E test-unit-sve_256 --rerun-failed --output-on-failure
+        # Disable SVE for now
+        run: ctest -E "test-unit-sve_128|test-unit-sve_256" --rerun-failed --output-on-failure
         working-directory: build_rel
 
   build-ubuntu-arm64-gcc:
@@ -78,13 +79,14 @@ jobs:
         with:
           submodules: 'true'
 
+      - name: Update apt packages
+        run: sudo apt-get update
+
       - name: Install ImageMagick
-        run: sudo apt install imagemagick
+        run: sudo apt-get install imagemagick
 
       - name: Install GCC 14
-        run: |
-          sudo apt-get update
-          sudo apt-get install -y gcc-14
+        run: sudo apt-get install -y gcc-14
 
       - name: Build release
         run: |
@@ -113,18 +115,16 @@ jobs:
           pip install numpy Pillow
 
       - name: Run system tests
-        # GitHub is using Cobalt-100, which has 128-bit SVE so we cannot test sve_256 here
+        # Disable SVE for now
         run: |
           python ./Test/astc_test_functional.py --encoder none
           python ./Test/astc_test_functional.py --encoder neon
-          python ./Test/astc_test_functional.py --encoder sve_128
           python ./Test/astc_test_image.py --encoder none --test-set Small
           python ./Test/astc_test_image.py --encoder neon --test-set Small
-          python ./Test/astc_test_image.py --encoder sve_128 --test-set Small
 
       - name: Run unit tests
-        # GitHub is using Cobalt-100, which has 128-bit SVE so we cannot test sve_256 here
-        run: ctest -E test-unit-sve_256 --rerun-failed --output-on-failure
+        # Disable SVE for now
+        run: ctest -E "test-unit-sve_128|test-unit-sve_256" --rerun-failed --output-on-failure
         working-directory: build_rel
 
   build-ubuntu-x64-clang:

external/astc-encoder/.github/workflows/post_weekly_release.yaml

@@ -71,8 +71,11 @@ jobs:
         with:
           submodules: 'true'
 
+      - name: Update apt packages
+        run: sudo apt-get update
+
       - name: Install ImageMagick
-        run: sudo apt install imagemagick
+        run: sudo apt-get install imagemagick
 
       - name: Build release
         run: |
@@ -101,12 +104,10 @@ jobs:
           pip install numpy Pillow
 
       - name: Run system tests
-        # GitHub is using Cobalt-100, which has 128-bit SVE so we cannot test sve_256 here
+        # Disable SVE testing for now
         run: |
           python ./Test/astc_test_functional.py --encoder neon
-          python ./Test/astc_test_functional.py --encoder sve_128
           python ./Test/astc_test_image.py --encoder neon --test-set Small
-          python ./Test/astc_test_image.py --encoder sve_128 --test-set Small
 
   build-ubuntu-x64:
     name: Ubuntu x64

external/astc-encoder/.gitignore

@@ -1,4 +1,5 @@
 # Editor and engineering scratch files
+.cache
 .vs
 .vscode
 .DS_Store

external/astc-encoder/.gitrepo

@@ -6,7 +6,7 @@
 [subrepo]
 	remote = https://github.com/ARM-software/astc-encoder.git
 	branch = 5.2.0
-	commit = 2319d9c4d4af53a7fc7c52985e264ce6e8a02a9b
-	parent = 7475bc1bc0411e672a0a833e19f1f53bde30b588
+	commit = 30aabb3f42406df45a910d8496f9bee17eeba9bb
+	parent = f9c73388a58de9b83f260f11008b043d8f7c0954
 	method = merge
-	cmdver = 0.4.3
+	cmdver = 0.4.9

external/astc-encoder/CMakeLists.txt

@@ -25,7 +25,7 @@ if(MSVC)
     add_compile_options("/wd4324")
 endif()
 
-project(astcencoder VERSION 5.2.0)
+project(astcencoder VERSION 5.3.0)
 
 set(CMAKE_CXX_STANDARD 14)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
@@ -45,13 +45,15 @@ option(ASTCENC_ISA_NONE "Enable astcenc builds for no SIMD")
 option(ASTCENC_ISA_NATIVE "Enable astcenc builds for native SIMD")
 option(ASTCENC_DECOMPRESSOR "Enable astcenc builds for decompression only")
 option(ASTCENC_SHAREDLIB "Enable astcenc builds with core library shared objects")
+option(ASTCENC_BIG_ENDIAN "Enable astcenc big-endian support")
 option(ASTCENC_DIAGNOSTICS "Enable astcenc builds with diagnostic trace")
 option(ASTCENC_ASAN "Enable astcenc builds with address sanitizer")
 option(ASTCENC_UBSAN "Enable astcenc builds with undefined behavior sanitizer")
 option(ASTCENC_UNITTEST "Enable astcenc builds with unit tests")
 option(ASTCENC_INVARIANCE "Enable astcenc floating point invariance" ON)
 option(ASTCENC_CLI "Enable build of astcenc command line tools" ON)
 option(ASTCENC_X86_GATHERS "Enable use of native x86 gathers" ON)
+option(ASTCENC_WERROR "Force builds to treat warnings as errors" ON)
 
 # Preflight for some macOS-specific build options
 if("${CMAKE_SYSTEM_NAME}" STREQUAL "Darwin")
@@ -137,6 +139,9 @@ if("${CMAKE_SYSTEM_NAME}" STREQUAL "Darwin")
     printopt("Universal bin      " ${ASTCENC_UNIVERSAL_BUILD})
 endif()
 printopt("Invariance         " ${ASTCENC_INVARIANCE})
+if(${ASTCENC_BIG_ENDIAN})
+    printopt("Big endian         " ${ASTCENC_BIG_ENDIAN})
+endif()
 printopt("Shared libs        " ${ASTCENC_SHAREDLIB})
 printopt("Decompressor       " ${ASTCENC_DECOMPRESSOR})
 message(STATUS "Developer options")

external/astc-encoder/Docs/Building-BE.md

@@ -0,0 +1,97 @@
+# Building Big Endian ASTC Encoder
+
+We don't officially support the big-endian compressor build, and it is only 
+lightly tested, but ensuring that the code is BE-compatible is good practice 
+and some of the open source distributions still support BE platforms. 
+
+Even though Arm64 can run in a BE mode, it's now very rare in practice. It's no
+longer supported out of the box in the latest Arm upstream compiler releases,
+and getting hold of a sysroot is increasingly difficult. To test BE builds, I
+therefore cross-compile Linux builds for PPC64 and use `qemu-user` to run
+them. This doesn't use a real sysroot, and so everything must be compiled with
+`-static` linkage.
+
+## Host software
+
+Install the following host software:
+
+```bash
+# Compiler
+sudo apt-get install g++-powerpc64-linux-gnu
+
+# Multi-arch libraries
+sudo apt-get install g++-multilib-powerpc64-linux-gnu
+
+# QEMU
+sudo apt-get install qemu-user-static qemu-user-binfmt binfmt-support
+sudo mkdir /etc/qemu-binfmt
+sudo ln -s /usr/powerpc64-linux-gnu /etc/qemu-binfmt/ppc64
+sudo update-binfmts --import qemu-ppc64
+```
+
+## CMake toolchain file
+
+Cross-compiling needs a correctly configured CMake, and the easiest way to
+do this consistently is to use a toolchain file. Create a `CMake-BE.toolchain`
+file in the root of the project, with the following content:
+
+```
+# Operating system
+set(CMAKE_SYSTEM_NAME Linux)
+
+# Cross-compilers for C and C++
+set(CMAKE_C_COMPILER   powerpc64-linux-gnu-gcc)
+set(CMAKE_CXX_COMPILER powerpc64-linux-gnu-g++)
+
+# Compiler environment
+set(CMAKE_FIND_ROOT_PATH /usr/powerpc64-linux-gnu)
+
+set(CMAKE_C_FLAGS_INIT -static)
+set(CMAKE_CXX_FLAGS_INIT -static)
+set(CMAKE_EXE_LINKER_FLAGS_INIT -static)
+set(CMAKE_SHARED_LINKER_FLAGS_INIT -static)
+set(CMAKE_MODULE_LINKER_FLAGS_INIT -static)
+
+# Build options
+set(ASTCENC_BIG_ENDIAN ON)
+
+# Never match host tools
+set(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
+
+# Only match headers and libraries in the compiler environment
+set(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)
+set(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)
+```
+
+## Build astcenc
+
+Building uses CMake as normal, with the additional specification of the
+toolchain file to configure the build for cross-compilation. We don't have any
+SIMD implementations for big-endian architectures so these builds must compile
+for the reference C SIMD implementation, `ASTCENC_ISA_NONE`.
+
+```
+cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=../ -DASTCENC_ISA_NONE=ON -DCMAKE_TOOLCHAIN_FILE=../CMake-BE.toolchain ..
+```
+
+## Run astcenc
+
+The cross-compiled `astcenc` binary runs as normal, and can access host files,
+but must run through QEMU to do the instruction set translation.
+
+If the binfmt setup performed earlier was successful you can just run the
+binary as if it were a native binary:
+
+```
+./bin/astcenc-none ...
+```
+
+... but otherwise you can run it manually using QEMU as a wrapper:
+
+```
+qemu-ppc64 ./bin/astcenc-none ...
+```
+
+- - -
+
+_Copyright © 2025, Arm Limited and contributors._

external/astc-encoder/Docs/ChangeLog-5x.md

@@ -6,6 +6,26 @@ release of the 5.x series.
 All performance data on this page is measured on an Intel Core i5-9600K
 clocked at 4.2 GHz, running `astcenc` using AVX2 and 6 threads.
 
+<!-- ---------------------------------------------------------------------- -->
+## 5.3.0
+
+**Status:** March 2025
+
+The 5.3.0 release is a minor maintenance release.
+
+* **General:**
+  * **Feature:** Reference C builds (`ASTCENC_ISA_NONE`) now support compiling
+    for big-endian CPUs. Compile with `-DASTCENC_BIG_ENDIAN=ON` when compiling
+    for a big-endian target; it is not auto-detected.
+  * **Improvement:** Builds using GCC now specify `-flto=auto` to allow
+    parallel link steps, and remove the log warnings about not setting a CPU
+    count parameter value.
+  * **Bug fix:** Builds using MSVC `cl.exe` that do not specify an explicit
+    ISA using the preprocessor configuration defines will now correctly
+    default to the SSE2 backend on x86-64 and the NEON backend on Arm64. Previously they would have defaulted to the reference C implementation,
+    which is around 3.25 times slower.
+
+
 <!-- ---------------------------------------------------------------------- -->
 ## 5.2.0
 

external/astc-encoder/README.md

@@ -77,10 +77,16 @@ The x86-64 SSE2 builds will work on all x86-64 machines, but it is the slowest
 of the three. The other two require extended CPU instruction set support which
 is not universally available, but each step gains ~15% more performance.
 
-For Arm, if binaries are available, we provide:
+For Arm we provide, in order of increasing performance:
 
+* `astcenc-sve_256` - uses 256-bit SVE
+* `astcenc-sve_128` - uses 128-bit SVE
 * `astcenc-neon` - uses NEON
 
+Note: The Arm Scalable Vector Extensions (SVE) allow CPUs to have a variable
+vector length. The astcenc implementation is not written in a length-agnostic
+style and requires the binary to match the vector length on the host CPU.
+
 ## macOS
 
 For macOS devices we provide a single universal binary `astcenc`, which allows
@@ -235,10 +241,10 @@ or general mobile graphics development or technology please submit them on the
 
 - - -
 
-_Copyright © 2013-2024, Arm Limited and contributors. All rights reserved._
+_Copyright © 2013-2025, Arm Limited and contributors. All rights reserved._
 
 [1]: ./Docs/FormatOverview.md
-[2]: https://www.khronos.org/registry/DataFormat/specs/1.3/dataformat.1.3.html#ASTC
+[2]: https://www.khronos.org/registry/DataFormat/specs/1.4/dataformat.1.4.html#ASTC
 [3]: https://github.com/ARM-software/astc-encoder/releases
 [4]: https://community.arm.com/support-forums/f/graphics-gaming-and-vr-forum/
 [5]: https://developer.arm.com/documentation/102162/latest/?lang=en

external/astc-encoder/Source/CMakeLists.txt

@@ -17,10 +17,14 @@
 
 # Overwrite the LTO flags to force fat LTO; worth 3-4% performance
 # See https://gitlab.kitware.com/cmake/cmake/-/issues/16808
-if(CMAKE_CXX_COMPILER_ID MATCHES "GNU|Clang" AND ${ASTCENC_CLI})
+if(CMAKE_CXX_COMPILER_ID MATCHES "Clang" AND ${ASTCENC_CLI})
     set(CMAKE_CXX_COMPILE_OPTIONS_IPO "-flto")
 endif()
 
+if(CMAKE_CXX_COMPILER_ID MATCHES "GNU" AND ${ASTCENC_CLI})
+    set(CMAKE_CXX_COMPILE_OPTIONS_IPO "-flto=auto")
+endif()
+
 if(${ASTCENC_DECOMPRESSOR})
     set(ASTCENC_CODEC dec)
 else()