ARROW-8440: [C++] Refine SIMD header files

cpp/cmake_modules/DefineOptions.cmake

@@ -95,9 +95,6 @@ if("${CMAKE_SOURCE_DIR}" STREQUAL "${CMAKE_CURRENT_SOURCE_DIR}")
   define_option(ARROW_USE_PRECOMPILED_HEADERS "Use precompiled headers when compiling"
                 OFF)
 
-  # Disable this option to exercise non-SIMD fallbacks
-  define_option(ARROW_USE_SIMD "Build with SIMD optimizations" ON)
-
   define_option_string(ARROW_SIMD_LEVEL
                        "SIMD compiler optimization level"
                        "SSE4_2" # default to SSE4.2

cpp/cmake_modules/SetupCxxFlags.cmake

@@ -305,7 +305,7 @@ if(BUILD_WARNING_FLAGS)
 endif(BUILD_WARNING_FLAGS)
 
 # Only enable additional instruction sets if they are supported
-if(ARROW_CPU_FLAG STREQUAL "x86" AND ARROW_USE_SIMD)
+if(ARROW_CPU_FLAG STREQUAL "x86")
   if(ARROW_SIMD_LEVEL STREQUAL "AVX512")
     if(NOT CXX_SUPPORTS_AVX512)
       message(FATAL_ERROR "AVX512 required but compiler doesn't support it.")
@@ -327,7 +327,7 @@ if(ARROW_CPU_FLAG STREQUAL "x86" AND ARROW_USE_SIMD)
   endif()
 endif()
 
-if(ARROW_CPU_FLAG STREQUAL "ppc" AND ARROW_USE_SIMD)
+if(ARROW_CPU_FLAG STREQUAL "ppc")
   if(CXX_SUPPORTS_ALTIVEC AND ARROW_ALTIVEC)
     set(CXX_COMMON_FLAGS "${CXX_COMMON_FLAGS} ${ARROW_ALTIVEC_FLAG}")
   endif()
@@ -342,9 +342,7 @@ if(ARROW_CPU_FLAG STREQUAL "armv8")
   endif()
   set(CXX_COMMON_FLAGS "${CXX_COMMON_FLAGS} ${ARROW_ARMV8_ARCH_FLAG}")
 
-  if(ARROW_USE_SIMD)
-    add_definitions(-DARROW_HAVE_NEON)
-  endif()
+  add_definitions(-DARROW_HAVE_NEON)
 
   if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU"
      AND CMAKE_CXX_COMPILER_VERSION VERSION_LESS "5.4")

cpp/src/arrow/io/memory_benchmark.cc

@@ -22,8 +22,7 @@
 #include "arrow/testing/gtest_util.h"
 #include "arrow/testing/util.h"
 #include "arrow/util/cpu_info.h"
-#include "arrow/util/neon_util.h"
-#include "arrow/util/sse_util.h"
+#include "arrow/util/simd.h"
 
 #include "benchmark/benchmark.h"
 
@@ -45,7 +44,7 @@ using BufferPtr = std::shared_ptr<Buffer>;
 
 #ifdef ARROW_HAVE_SSE4_2
 
-#ifdef ARROW_AVX512
+#ifdef ARROW_HAVE_AVX512
 
 using VectorType = __m512i;
 #define VectorSet _mm512_set1_epi32
@@ -59,7 +58,7 @@ using VectorType = __m512i;
 
 #else
 
-#ifdef ARROW_AVX2
+#ifdef ARROW_HAVE_AVX2
 
 using VectorType = __m256i;
 #define VectorSet _mm256_set1_epi32
@@ -71,7 +70,7 @@ using VectorType = __m256i;
   asm volatile("vmovntdqa %[src], %[dst]" : [ dst ] "=v"(DST) : [ src ] "m"(SRC) :)
 #define VectorStreamWrite _mm256_stream_si256
 
-#else  // ARROW_AVX2 not set
+#else  // ARROW_HAVE_AVX2 not set
 
 using VectorType = __m128i;
 #define VectorSet _mm_set1_epi32
@@ -83,8 +82,8 @@ using VectorType = __m128i;
   asm volatile("movntdqa %[src], %[dst]" : [ dst ] "=x"(DST) : [ src ] "m"(SRC) :)
 #define VectorStreamWrite _mm_stream_si128
 
-#endif  // ARROW_AVX2
-#endif  // ARROW_AVX512
+#endif  // ARROW_HAVE_AVX2
+#endif  // ARROW_HAVE_AVX512
 
 static void Read(void* src, void* dst, size_t size) {
   const auto simd = static_cast<VectorType*>(src);

cpp/src/arrow/json/rapidjson_defs.h

@@ -32,9 +32,6 @@
   }                             \
   }
 
-#include "arrow/util/neon_util.h"
-#include "arrow/util/sse_util.h"
-
 // enable SIMD whitespace skipping, if available
 #if defined(ARROW_HAVE_SSE4_2)
 #define RAPIDJSON_SSE2 1

cpp/src/arrow/util/byte_stream_split.h

@@ -17,16 +17,12 @@
 
 #pragma once
 
-#include "arrow/util/sse_util.h"
+#include "arrow/util/simd.h"
 #include "arrow/util/ubsan.h"
 
 #include <stdint.h>
 #include <algorithm>
 
-#ifdef ARROW_HAVE_AVX2
-#include <immintrin.h>
-#endif  // ARROW_HAVE_AVX2
-
 #ifdef ARROW_HAVE_SSE4_2
 // Enable the SIMD for ByteStreamSplit Encoder/Decoder
 #define ARROW_HAVE_SIMD_SPLIT