Add support for fixed-width 128-bit SVE implementation

CMakeLists.txt

@@ -39,6 +39,7 @@ option(ASTCENC_ISA_AVX2 "Enable astcenc builds for AVX2 SIMD")
 option(ASTCENC_ISA_SSE41 "Enable astcenc builds for SSE4.1 SIMD")
 option(ASTCENC_ISA_SSE2 "Enable astcenc builds for SSE2 SIMD")
 option(ASTCENC_ISA_SVE_256 "Enable astcenc builds for 256-bit SVE SIMD")
+option(ASTCENC_ISA_SVE_128 "Enable astcenc builds for 128-bit SVE SIMD")
 option(ASTCENC_ISA_NEON "Enable astcenc builds for NEON SIMD")
 option(ASTCENC_ISA_NONE "Enable astcenc builds for no SIMD")
 option(ASTCENC_ISA_NATIVE "Enable astcenc builds for native SIMD")
@@ -87,7 +88,7 @@ endforeach()
 
 # Count options which MUST be arm64
 set(ASTCENC_ARM64_ISA_COUNT 0)
-set(ASTCENC_CONFIGS ${ASTCENC_ISA_NEON} ${ASTCENC_ISA_SVE_256})
+set(ASTCENC_CONFIGS ${ASTCENC_ISA_NEON} ${ASTCENC_ISA_SVE_128} ${ASTCENC_ISA_SVE_256})
 foreach(ASTCENC_CONFIG ${ASTCENC_CONFIGS})
     if(${ASTCENC_CONFIG})
         math(EXPR ASTCENC_ARM64_ISA_COUNT "${ASTCENC_ARM64_ISA_COUNT} + 1")
@@ -120,6 +121,7 @@ endif()
 
 message(STATUS "Arm backend options")
 printopt("SVE 256b backend   " ${ASTCENC_ISA_SVE_256})
+printopt("SVE 128b backend   " ${ASTCENC_ISA_SVE_128})
 printopt("NEON backend       " ${ASTCENC_ISA_NEON})
 message(STATUS "x86-64 backend options")
 printopt("AVX2 backend       " ${ASTCENC_ISA_AVX2})

Docs/ChangeLog-4x.md

@@ -18,6 +18,7 @@ The 4.9.0 release is a minor maintenance release.
     reference implementation.
   * **Bug fix:** Fixed sincos table index under/overflow.
   * **Feature:** Added backend for Arm SVE fixed-width 256-bit builds.
+  * **Feature:** Added backend for Arm SVE fixed-width 128-bit builds.
   * **Feature:** Optimized NEON mask `any()` and `all()` functions.
   * **Feature:** Migrated build and test to GitHub Actions pipelines.
 

Source/CMakeLists.txt

@@ -27,8 +27,8 @@ else()
     set(ASTCENC_CODEC enc)
 endif()
 
-set(ASTCENC_ARTIFACTS native none sve_256 neon avx2 sse4.1 sse2)
-set(ASTCENC_CONFIGS ${ASTCENC_ISA_NATIVE} ${ASTCENC_ISA_NONE} ${ASTCENC_ISA_SVE_256} ${ASTCENC_ISA_NEON} ${ASTCENC_ISA_AVX2} ${ASTCENC_ISA_SSE41} ${ASTCENC_ISA_SSE2})
+set(ASTCENC_ARTIFACTS native none sve_256 sve_128 neon avx2 sse4.1 sse2)
+set(ASTCENC_CONFIGS ${ASTCENC_ISA_NATIVE} ${ASTCENC_ISA_NONE} ${ASTCENC_ISA_SVE_256} ${ASTCENC_ISA_SVE_128} ${ASTCENC_ISA_NEON} ${ASTCENC_ISA_AVX2} ${ASTCENC_ISA_SSE41} ${ASTCENC_ISA_SSE2})
 list(LENGTH ASTCENC_ARTIFACTS ASTCENC_ARTIFACTS_LEN)
 math(EXPR ASTCENC_ARTIFACTS_LEN "${ASTCENC_ARTIFACTS_LEN} - 1")
 
@@ -40,6 +40,8 @@ foreach(INDEX RANGE ${ASTCENC_ARTIFACTS_LEN})
 
         if(${ASTCENC_ISA_SIMD} MATCHES "sve_256")
            # Not suported on macOS
+        elseif(${ASTCENC_ISA_SIMD} MATCHES "sve_128")
+           # Not suported on macOS
         elseif(${ASTCENC_ISA_SIMD} MATCHES "neon")
            set(CMAKE_OSX_ARCHITECTURES arm64)
         elseif(${ASTCENC_ISA_SIMD} MATCHES "sse2")

Source/UnitTest/CMakeLists.txt

@@ -15,8 +15,8 @@
 #  under the License.
 #  ----------------------------------------------------------------------------
 
-set(ASTCENC_ARTIFACTS native none sve_256 neon avx2 sse4.1 sse2)
-set(ASTCENC_CONFIGS ${ASTCENC_ISA_NATIVE} ${ASTCENC_ISA_NONE} ${ASTCENC_ISA_SVE_256} ${ASTCENC_ISA_NEON} ${ASTCENC_ISA_AVX2} ${ASTCENC_ISA_SSE41} ${ASTCENC_ISA_SSE2})
+set(ASTCENC_ARTIFACTS native none sve_256 sve_128 neon avx2 sse4.1 sse2)
+set(ASTCENC_CONFIGS ${ASTCENC_ISA_NATIVE} ${ASTCENC_ISA_NONE} ${ASTCENC_ISA_SVE_256} ${ASTCENC_ISA_SVE_128} ${ASTCENC_ISA_NEON} ${ASTCENC_ISA_AVX2} ${ASTCENC_ISA_SSE41} ${ASTCENC_ISA_SSE2})
 list(LENGTH ASTCENC_ARTIFACTS ASTCENC_ARTIFACTS_LEN)
 math(EXPR ASTCENC_ARTIFACTS_LEN "${ASTCENC_ARTIFACTS_LEN} - 1")
 
@@ -28,6 +28,8 @@ foreach(INDEX RANGE ${ASTCENC_ARTIFACTS_LEN})
 
         if(${ASTCENC_ISA_SIMD} MATCHES "sve_256")
            # Not supported on macOS
+        elseif(${ASTCENC_ISA_SIMD} MATCHES "sve_128")
+           # Not supported on macOS
         elseif(${ASTCENC_ISA_SIMD} MATCHES "neon")
            set(CMAKE_OSX_ARCHITECTURES arm64)
         elseif(${ASTCENC_ISA_SIMD} MATCHES "sse2")

Source/UnitTest/cmake_core.cmake

@@ -104,6 +104,21 @@ elseif(${ASTCENC_ISA_SIMD} MATCHES "sve_256")
         PRIVATE
             -march=armv8-a+sve -msve-vector-bits=256)
 
+elseif(${ASTCENC_ISA_SIMD} MATCHES "sve_128")
+    target_compile_definitions(${ASTCENC_TEST}
+        PRIVATE
+            ASTCENC_NEON=1
+            ASTCENC_SVE=4
+            ASTCENC_SSE=0
+            ASTCENC_AVX=0
+            ASTCENC_POPCNT=0
+            ASTCENC_F16C=0)
+
+    # Enable SVE
+    target_compile_options(${ASTCENC_TEST}
+        PRIVATE
+            -march=armv8-a+sve -msve-vector-bits=128)
+
 elseif(${ASTCENC_ISA_SIMD} MATCHES "sse2")
     target_compile_definitions(${ASTCENC_TEST}
         PRIVATE

Source/astcenccli_toplevel_help.cpp

@@ -562,6 +562,8 @@ void astcenc_print_header()
 	const char* simdtype = "sse2";
 #elif (ASTCENC_SVE == 8)
 	const char* simdtype = "sve.256b";
+#elif (ASTCENC_SVE == 4)
+	const char* simdtype = "sve.128b";
 #elif (ASTCENC_NEON == 1)
 	const char* simdtype = "neon";
 #else

Source/cmake_core.cmake

@@ -325,6 +325,29 @@ macro(astcenc_set_properties ASTCENC_TARGET_NAME ASTCENC_VENEER_TYPE)
                     -march=armv8-a+sve)
         endif()
 
+    elseif(${ASTCENC_ISA_SIMD} MATCHES "sve_128")
+        target_compile_definitions(${ASTCENC_TARGET_NAME}
+            PRIVATE
+                ASTCENC_NEON=1
+                ASTCENC_SVE=4
+                ASTCENC_SSE=0
+                ASTCENC_AVX=0
+                ASTCENC_POPCNT=0
+                ASTCENC_F16C=0)
+
+        # Enable SVE in the core library
+        if (NOT ${ASTCENC_VENEER_TYPE})
+            target_compile_options(${ASTCENC_TARGET_NAME}
+                PRIVATE
+                    -march=armv8-a+sve -msve-vector-bits=128)
+
+        # Enable SVE without fixed vector length in the veneer
+        elseif (${ASTCENC_VENEER_TYPE} EQUAL 2)
+            target_compile_options(${ASTCENC_TARGET_NAME}
+                PRIVATE
+                    -march=armv8-a+sve)
+        endif()
+
     elseif(${ASTCENC_ISA_SIMD} MATCHES "sse2")
         target_compile_definitions(${ASTCENC_TARGET_NAME}
             PRIVATE

Test/astc_test_image.py

@@ -306,11 +306,11 @@ def parse_command_line():
                  "ref-2.5-neon", "ref-2.5-sse2", "ref-2.5-sse4.1", "ref-2.5-avx2",
                  "ref-3.7-neon", "ref-3.7-sse2", "ref-3.7-sse4.1", "ref-3.7-avx2",
                  "ref-4.8-neon", "ref-4.8-sse2", "ref-4.8-sse4.1", "ref-4.8-avx2",
-                 "ref-main-neon", "ref-main-sve_256", "ref-main-sse2", "ref-main-sse4.1", "ref-main-avx2"]
+                 "ref-main-neon", "ref-main-sve_256", "ref-main-sve_128", "ref-main-sse2", "ref-main-sse4.1", "ref-main-avx2"]
 
     # All test encoders
-    testcoders = ["none", "neon", "sve_256", "sse2", "sse4.1", "avx2", "native", "universal"]
-    testcodersAArch64 = ["neon", "sve_256"]
+    testcoders = ["none", "neon", "sve_256", "sve_128", "sse2", "sse4.1", "avx2", "native", "universal"]
+    testcodersAArch64 = ["neon", "sve_256", "sve_128"]
     testcodersX86 = ["sse2", "sse4.1", "avx2"]
 
     coders = refcoders + testcoders + ["all-aarch64", "all-x86"]