Add SVE, SVE2, and SVE2_BITPERM as targets

Change-Id: I5231e2eb0a31708a16c853dc83ea48db32e0b0a5

diff --git a/CMakeLists.txt b/CMakeLists.txt
index e32be7b30f9e1f9a6f7283a378766f77ea5ca115..fa9648f7a6d4de4443001f7f0df7e7760b69da72 100644 (file)
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -213,6 +213,14 @@ else()
         set(TUNE_FLAG native)
     endif()
 
+    if (BUILD_SVE2_BITPERM)
+        set(GNUCC_ARCH "${GNUCC_ARCH}+sve2-bitperm")
+    elseif (BUILD_SVE2)
+        set(GNUCC_ARCH "${GNUCC_ARCH}+sve2")
+    elseif (BUILD_SVE)
+        set(GNUCC_ARCH "${GNUCC_ARCH}+sve")
+    endif ()
+
     # compiler version checks TODO: test more compilers
     if (CMAKE_COMPILER_IS_GNUCXX)
         set(GNUCXX_MINVER "4.8.1")
@@ -296,6 +304,9 @@ if (ARCH_IA32 OR ARCH_X86_64)
   CHECK_INCLUDE_FILE_CXX(x86intrin.h HAVE_CXX_X86INTRIN_H)
 elseif (ARCH_ARM32 OR ARCH_AARCH64)
   CHECK_INCLUDE_FILE_CXX(arm_neon.h HAVE_C_ARM_NEON_H)
+  if (BUILD_SVE OR BUILD_SVE2 OR BUILD_SVE2_BITPERM)
+    CHECK_INCLUDE_FILE_CXX(arm_sve.h HAVE_C_ARM_SVE_H)
+  endif()
   set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -flax-vector-conversions")
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -flax-vector-conversions")
 endif()

diff --git a/README.md b/README.md
index e780238f651c88da8f83d0d013d3af32f89ec726..8bc7aff64540df34d73c19cccd72debcff40988b 100644 (file)
--- a/README.md
+++ b/README.md
@@ -45,6 +45,18 @@ Vectorscan is typically used in a DPI library stack, just like Hyperscan.
   - `make -jT` where T is the number of threads used to compile.
   - `cmake --build . -- -j T` can also be used instead of make.
 
+# Compiling for SVE
+
+The following cmake variables can be set in order to target Arm's Scalable
+Vector Extension. They are listed in ascending order of strength, with cmake
+detecting whether the feature is available in the compiler and falling back to
+a weaker version if not. Only one of these variables needs to be set as weaker
+variables will be implied as set.
+
+- `BUILD_SVE`
+- `BUILD_SVE2`
+- `BUILD_SVE2_BITPERM`
+
 # Documentation
 
 Information on building the Hyperscan library and using its API is available in

diff --git a/cmake/arch.cmake b/cmake/arch.cmake
index 691861d66338166b2be958a10ba076f9e23d297f..c757e91cee5f7f20c256e883b747e598b33faa4a 100644 (file)
--- a/cmake/arch.cmake
+++ b/cmake/arch.cmake
@@ -13,6 +13,52 @@ else()
     message (FATAL_ERROR "No intrinsics header found")
 endif ()
 
+if (ARCH_ARM32 OR ARCH_AARCH64)
+    CHECK_C_SOURCE_COMPILES("#include <${INTRIN_INC_H}>
+int main() {
+    int32x4_t a = vdupq_n_s32(1);
+    (void)a;
+}" HAVE_NEON)
+endif ()
+
+if (ARCH_AARCH64)
+    set(PREV_FLAGS "${CMAKE_C_FLAGS}")
+    if (BUILD_SVE2_BITPERM)
+        set(CMAKE_C_FLAGS "-march=${GNUCC_ARCH} ${CMAKE_C_FLAGS}")
+        CHECK_C_SOURCE_COMPILES("#include <arm_sve.h>
+        int main() {
+            svuint8_t a = svbext(svdup_u8(1), svdup_u8(2));
+            (void)a;
+        }" HAVE_SVE2_BITPERM)
+        if (HAVE_SVE2_BITPERM)
+            add_definitions(-DHAVE_SVE2_BITPERM)
+        endif ()
+    endif()
+    if (BUILD_SVE2)
+        set(CMAKE_C_FLAGS "-march=${GNUCC_ARCH} ${CMAKE_C_FLAGS}")
+        CHECK_C_SOURCE_COMPILES("#include <arm_sve.h>
+        int main() {
+            svuint8_t a = svbsl(svdup_u8(1), svdup_u8(2), svdup_u8(3));
+            (void)a;
+        }" HAVE_SVE2)
+        if (HAVE_SVE2)
+            add_definitions(-DHAVE_SVE2)
+        endif ()
+    endif()
+    if (BUILD_SVE)
+        set(CMAKE_C_FLAGS "-march=${GNUCC_ARCH} ${CMAKE_C_FLAGS}")
+        CHECK_C_SOURCE_COMPILES("#include <arm_sve.h>
+        int main() {
+            svuint8_t a = svdup_u8(1);
+            (void)a;
+        }" HAVE_SVE)
+        if (HAVE_SVE)
+            add_definitions(-DHAVE_SVE)
+        endif ()
+    endif ()
+    set(CMAKE_C_FLAGS "${PREV_FLAGS}")
+endif()
+
 if (BUILD_AVX512)
     CHECK_C_COMPILER_FLAG(${SKYLAKE_FLAG} HAS_ARCH_SKYLAKE)
     if (NOT HAS_ARCH_SKYLAKE)
@@ -90,13 +136,7 @@ int main(){
     (void)_mm512_permutexvar_epi8(idx, a);
 }" HAVE_AVX512VBMI)
 
-elseif (ARCH_ARM32 OR ARCH_AARCH64)
-    CHECK_C_SOURCE_COMPILES("#include <${INTRIN_INC_H}>
-int main() {
-    int32x4_t a = vdupq_n_s32(1);
-    (void)a;
-}" HAVE_NEON)
-else ()
+elseif (!ARCH_ARM32 AND !ARCH_AARCH64)
     message (FATAL_ERROR "Unsupported architecture")
 endif ()
 
@@ -131,5 +171,6 @@ else (NOT FAT_RUNTIME)
     endif ()
 endif ()
 
+unset (PREV_FLAGS)
 unset (CMAKE_REQUIRED_FLAGS)
 unset (INTRIN_INC_H)

diff --git a/cmake/config.h.in b/cmake/config.h.in
index 17c1e72933034f6dc9b21bfec02e500be9eecc2a..0afd6998cdfa7298263b79168c628128cae5fd79 100644 (file)
--- a/cmake/config.h.in
+++ b/cmake/config.h.in
@@ -24,6 +24,15 @@
 /* "Define if cross compiling for AARCH64" */
 #cmakedefine CROSS_COMPILE_AARCH64
 
+/* Define if building SVE for AARCH64. */
+#cmakedefine BUILD_SVE
+
+/* Define if building SVE2 for AARCH64. */
+#cmakedefine BUILD_SVE2
+
+/* Define if building SVE2+BITPERM for AARCH64. */
+#cmakedefine BUILD_SVE2_BITPERM
+
 /* internal build, switch on dump support. */
 #cmakedefine DUMP_SUPPORT
 
@@ -63,6 +72,9 @@
 /* C compiler has arm_neon.h */
 #cmakedefine HAVE_C_ARM_NEON_H
 
+/* C compiler has arm_sve.h */
+#cmakedefine HAVE_C_ARM_SVE_H
+
 /* Define to 1 if you have the declaration of `pthread_setaffinity_np', and to
    0 if you don't. */
 #cmakedefine HAVE_DECL_PTHREAD_SETAFFINITY_NP

diff --git a/src/util/intrinsics.h b/src/util/intrinsics.h
index 3e2afc22450cc7aac5615f9cba16419e905d53f5..33beb49755623e06b29232bb3e8e12643a3b499f 100644 (file)
--- a/src/util/intrinsics.h
+++ b/src/util/intrinsics.h
@@ -47,6 +47,15 @@
 
 #if defined(HAVE_C_ARM_NEON_H)
 #  define USE_ARM_NEON_H
+#  if defined(HAVE_C_ARM_SVE_H)
+#    define USE_ARM_SVE
+#    if defined(BUILD_SVE2)
+#      define USE_ARM_SVE2
+#      if defined(BUILD_SVE2_BITPERM)
+#        define USE_ARM_SVE2_BITPERM
+#      endif
+#    endif
+#  endif
 #endif
 
 #ifdef __cplusplus
@@ -65,6 +74,9 @@
 #include <intrin.h>
 #elif defined(USE_ARM_NEON_H)
 #include <arm_neon.h>
+#  if defined(USE_ARM_SVE)
+#    include <arm_sve.h>
+#  endif
 #else
 #error no intrinsics file
 #endif