Merge pull request #474 from georges-arm/geoste01/fastFwdCore_2D-sve

Add AArch64 SVE implementation for TCoeffOps fastFwdCore_2D

CMakeLists.txt

@@ -32,9 +32,12 @@ if( VVENC_TARGET_ARCH MATCHES "ARM\|AARCH64" )
     set( CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fno-lax-vector-conversions" )
     set( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-lax-vector-conversions" )
   endif()
+
+  # Check if SVE/SVE2 flags are supported by the compiler.
+  set_if_compiler_supports_arm_extensions( FLAG_sve FLAG_sve2 )
 endif()
 
-# we enable x86 intrinsics for all target architectures, because they are implemented through simd-everywhere on non-x86
+# We enable x86 intrinsics for all target architectures, because they are implemented through SIMD-everywhere on non-x86.
 set( VVENC_ENABLE_X86_SIMD TRUE                      CACHE BOOL "Enable x86 intrinsics" )
 set( VVENC_ENABLE_ARM_SIMD ${VVENC_ARM_SIMD_DEFAULT} CACHE BOOL "Enable Arm Neon intrinsics" )
 set( VVENC_ENABLE_ARM_SIMD_SVE FALSE                 CACHE BOOL "Enable Arm SVE intrinsics" )
@@ -81,27 +84,25 @@ if( VVENC_TARGET_ARCH STREQUAL "AARCH64" )
   if( VVENC_ENABLE_ARM_SIMD )
     message( STATUS "AArch64 Neon intrinsics enabled" )
     add_compile_definitions( TARGET_SIMD_ARM=1 )
-  else()
-    message( STATUS "AArch64 Neon intrinsics disabled, disabling AArch64 SVE/SVE2 intrinsics" )
-    # If Neon is disabled make sure that SVE/SVE2 are also disabled.
-    set( VVENC_ENABLE_ARM_SIMD_SVE FALSE )
-    set( VVENC_ENABLE_ARM_SIMD_SVE2 FALSE )
   endif()
 
-  if( VVENC_ENABLE_ARM_SIMD_SVE )
+  # If Neon is disabled or SVE is not supported, make sure that SVE is disabled.
+  if ( NOT VVENC_ENABLE_ARM_SIMD OR NOT FLAG_sve )
+    message( STATUS "Disabling AArch64 SVE/SVE2 intrinsics" )
+    set( VVENC_ENABLE_ARM_SIMD_SVE FALSE )
+  elseif( VVENC_ENABLE_ARM_SIMD_SVE )
     message( STATUS "AArch64 SVE intrinsics enabled" )
     add_compile_definitions( TARGET_SIMD_ARM_SVE=1 )
-  else()
-    message( STATUS "AArch64 SVE intrinsics disabled, disabling AArch64 SVE2 intrinsics" )
-    # If SVE is disabled make sure that SVE2 are also disabled.
-    set( VVENC_ENABLE_ARM_SIMD_SVE2 FALSE )
   endif()
 
-  if( VVENC_ENABLE_ARM_SIMD_SVE2 )
+  # If SVE is disabled or SVE2 is not supported, make sure that SVE2 is disabled.
+  if ( NOT VVENC_ENABLE_ARM_SIMD_SVE OR NOT FLAG_sve2 )
+    message( STATUS "Disabling AArch64 SVE2 intrinsics" )
+    # If SVE is disabled make sure that SVE2 are also disabled.
+    set( VVENC_ENABLE_ARM_SIMD_SVE2 FALSE )
+  elseif( VVENC_ENABLE_ARM_SIMD_SVE2 )
     message( STATUS "AArch64 SVE2 intrinsics enabled" )
     add_compile_definitions( TARGET_SIMD_ARM_SVE2=1 )
-  else()
-    message( STATUS "AArch64 SVE2 intrinsics disabled" )
   endif()
 endif()
 

cmake/modules/vvencCompilerSupport.cmake

@@ -1,5 +1,6 @@
 include( CheckCCompilerFlag )
 include( CheckCSourceCompiles )
+include( CheckCXXSourceCompiles )
 
 function( set_if_compiler_supports_flag output_var flag )
   string( REGEX REPLACE "[-.]" "_" SUPPORTED_flag_var "SUPPORTED${flag}" )
@@ -84,6 +85,71 @@ function( _emscripten_enable_wasm_simd128 )
   endif()
 endfunction()
 
+function( _set_if_compiler_supports_sve_flag output_var sve_flag )
+  set_if_compiler_supports_flag( tmp_var "${sve_flag}" )
+  if( NOT tmp_var )
+    return()
+  endif()
+
+  set( OLD_CMAKE_REQUIRED_FLAGS ${CMAKE_REQUIRED_FLAGS} )
+  set( OLD_CMAKE_TRY_COMPILE_TARGET_TYPE ${CMAKE_TRY_COMPILE_TARGET_TYPE} )
+  set( CMAKE_REQUIRED_FLAGS "${CMAKE_REQUIRED_FLAGS} ${sve_flag}" )
+  set( CMAKE_TRY_COMPILE_TARGET_TYPE STATIC_LIBRARY )
+
+  # Check whether the compiler can compile SVE functions that require
+  # backup/restore of SVE registers according to AAPCS.
+  # https://github.com/llvm/llvm-project/issues/80009.
+  set( SVE_COMPILATION_TEST "
+#include <arm_sve.h>
+void other();
+svfloat32_t func(svfloat32_t a) {
+other();
+return a;
+}
+int main() { return 0; }" )
+
+  check_c_source_compiles( "${SVE_COMPILATION_TEST}" SVE_COMPILATION_C_TEST_COMPILED )
+  check_cxx_source_compiles( "${SVE_COMPILATION_TEST}" SVE_COMPILATION_CXX_TEST_COMPILED )
+
+  # Check if arm_neon_sve_bridge.h is available.
+  set( SVE_HEADER_TEST "
+#ifndef __ARM_NEON_SVE_BRIDGE
+#error 1
+#endif
+#include <arm_sve.h>
+#include <arm_neon_sve_bridge.h>
+int main() { return 0; }")
+  check_c_source_compiles( "${SVE_HEADER_TEST}" SVE_HEADER_C_TEST_COMPILED )
+  check_cxx_source_compiles( "${SVE_HEADER_TEST}" SVE_HEADER_CXX_TEST_COMPILED )
+
+  set( CMAKE_REQUIRED_FLAGS ${OLD_CMAKE_REQUIRED_FLAGS} )
+  set( CMAKE_TRY_COMPILE_TARGET_TYPE ${OLD_CMAKE_TRY_COMPILE_TARGET_TYPE} )
+
+  if( SVE_COMPILATION_C_TEST_COMPILED AND SVE_COMPILATION_CXX_TEST_COMPILED AND
+      SVE_HEADER_C_TEST_COMPILED AND SVE_HEADER_CXX_TEST_COMPILED )
+    set( ${output_var} "${tmp_var}" PARENT_SCOPE )
+  endif()
+endfunction()
+
+# Check if the compiler supports the AArch64 SVE and SVE2 extensions, and set
+# variables for flags used to enable them to avoid duplication.
+function( set_if_compiler_supports_arm_extensions output_flag_sve output_flag_sve2 )
+  if( NOT(( ${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm64" ) OR
+          ( ${CMAKE_SYSTEM_PROCESSOR} MATCHES "aarch64" )))
+    return()
+  endif()
+  if( UNIX OR MINGW )
+    # SVE is an optional feature from Armv8.2-A.
+    set( _flag_sve "-march=armv8.2-a+sve" )
+    _set_if_compiler_supports_sve_flag( _sve_supported "${_flag_sve}" )
+    set( ${output_flag_sve} "${_sve_supported}" PARENT_SCOPE )
+
+    # SVE2 is an optional feature from Armv9.0-A.
+    set( _flag_sve2 "-march=armv9-a+sve2" )
+    _set_if_compiler_supports_sve_flag( _sve2_supported "${_flag_sve2}" )
+    set( ${output_flag_sve2} "${_sve2_supported}" PARENT_SCOPE )
+  endif()
+endfunction()
 
 function( check_problematic_compiler output_var compiler_id first_bad_version first_fixed_version )
   if( CMAKE_CXX_COMPILER_ID STREQUAL "${compiler_id}"

source/Lib/CommonLib/arm/InitARM.cpp

@@ -118,8 +118,14 @@ void TCoeffOps::initTCoeffOpsARM()
   {
     _initTCoeffOpsARM<NEON>();
   }
+#if TARGET_SIMD_ARM_SVE
+  if( vext >= SVE )
+  {
+    _initTCoeffOpsARM<SVE>();
+  }
+#endif  // TARGET_SIMD_ARM_SVE
 }
-#endif
+#endif  // ENABLE_SIMD_TRAFO
 
 #if ENABLE_SIMD_OPT_BDOF
 void InterPredInterpolation::initInterPredictionARM()
@@ -135,8 +141,6 @@ void InterPredInterpolation::initInterPredictionARM()
 }
 #endif
 
-
-
-#endif   // TARGET_SIMD_ARM
+#endif  // TARGET_SIMD_ARM
 
 }   // namespace