cpu.h: add WEBP_AARCH64

and define it to true for __aarch64__ and Win Arm64 + Visual Studio.

Microsoft's compiler (cl.exe) does not define __aarch64__, but relies on
_M_ARM64 & _M_ARM64EC

Bug: b/277254922
Change-Id: I20e4fa07a4031599db69e3d7ba9050345315ef51

src/dec/tree_dec.c

@@ -12,10 +12,11 @@
 // Author: Skal (pascal.massimino@gmail.com)
 
 #include "src/dec/vp8i_dec.h"
+#include "src/dsp/cpu.h"
 #include "src/utils/bit_reader_inl_utils.h"
 
 #if !defined(USE_GENERIC_TREE)
-#if !defined(__arm__) && !defined(_M_ARM) && !defined(__aarch64__)
+#if !defined(__arm__) && !defined(_M_ARM) && !WEBP_AARCH64
 // using a table is ~1-2% slower on ARM. Prefer the coded-tree approach then.
 #define USE_GENERIC_TREE 1   // ALTERNATE_CODE
 #else

src/dsp/cost_neon.c

@@ -29,7 +29,7 @@ static void SetResidualCoeffs_NEON(const int16_t* const coeffs,
   const uint8x16_t eob = vcombine_u8(vqmovn_u16(eob_0), vqmovn_u16(eob_1));
   const uint8x16_t masked = vandq_u8(eob, vld1q_u8(position));
 
-#ifdef __aarch64__
+#if WEBP_AARCH64
   res->last = vmaxvq_u8(masked) - 1;
 #else
   const uint8x8_t eob_8x8 = vmax_u8(vget_low_u8(masked), vget_high_u8(masked));
@@ -43,7 +43,7 @@ static void SetResidualCoeffs_NEON(const int16_t* const coeffs,
 
   vst1_lane_s32(&res->last, vreinterpret_s32_u32(eob_32x2), 0);
   --res->last;
-#endif  // __aarch64__
+#endif  // WEBP_AARCH64
 
   res->coeffs = coeffs;
 }

src/dsp/cpu.h

@@ -105,6 +105,12 @@
 #define WEBP_USE_INTRINSICS
 #endif
 
+#if defined(__aarch64__) || defined(_M_ARM64) || defined(_M_ARM64EC)
+#define WEBP_AARCH64 1
+#else
+#define WEBP_AARCH64 0
+#endif
+
 #if defined(WEBP_USE_NEON) && !defined(WEBP_HAVE_NEON)
 #define WEBP_HAVE_NEON
 #endif
@@ -134,8 +140,7 @@
 #define WEBP_NEON_OMIT_C_CODE 0
 #endif
 
-#if !(LOCAL_CLANG_PREREQ(3, 8) || LOCAL_GCC_PREREQ(4, 8) || \
+#if !(LOCAL_CLANG_PREREQ(3, 8) || LOCAL_GCC_PREREQ(4, 8) || WEBP_AARCH64)
-      defined(__aarch64__))
 #define WEBP_NEON_WORK_AROUND_GCC 1
 #else
 #define WEBP_NEON_WORK_AROUND_GCC 0

src/dsp/dec_neon.c

@@ -1428,7 +1428,7 @@ static WEBP_INLINE void DC8_NEON(uint8_t* dst, int do_top, int do_left) {
 
   if (do_top) {
     const uint8x8_t A = vld1_u8(dst - BPS);  // top row
-#if defined(__aarch64__)
+#if WEBP_AARCH64
     const uint16_t p2 = vaddlv_u8(A);
     sum_top = vdupq_n_u16(p2);
 #else
@@ -1511,7 +1511,7 @@ static WEBP_INLINE void DC16_NEON(uint8_t* dst, int do_top, int do_left) {
 
   if (do_top) {
     const uint8x16_t A = vld1q_u8(dst - BPS);  // top row
-#if defined(__aarch64__)
+#if WEBP_AARCH64
     const uint16_t p3 = vaddlvq_u8(A);
     sum_top = vdupq_n_u16(p3);
 #else

src/dsp/enc_neon.c

@@ -764,7 +764,7 @@ static WEBP_INLINE void AccumulateSSE16_NEON(const uint8_t* const a,
 
 // Horizontal sum of all four uint32_t values in 'sum'.
 static int SumToInt_NEON(uint32x4_t sum) {
-#if defined(__aarch64__)
+#if WEBP_AARCH64
   return (int)vaddvq_u32(sum);
 #else
   const uint64x2_t sum2 = vpaddlq_u32(sum);
@@ -865,7 +865,7 @@ static int QuantizeBlock_NEON(int16_t in[16], int16_t out[16],
   uint8x8x4_t shuffles;
   // vtbl?_u8 are marked unavailable for iOS arm64 with Xcode < 6.3, use
   // non-standard versions there.
-#if defined(__APPLE__) && defined(__aarch64__) && \
+#if defined(__APPLE__) && WEBP_AARCH64 && \
     defined(__apple_build_version__) && (__apple_build_version__< 6020037)
   uint8x16x2_t all_out;
   INIT_VECTOR2(all_out, vreinterpretq_u8_s16(out0), vreinterpretq_u8_s16(out1));

src/dsp/lossless_enc_neon.c

@@ -25,7 +25,7 @@
 
 // vtbl?_u8 are marked unavailable for iOS arm64 with Xcode < 6.3, use
 // non-standard versions there.
-#if defined(__APPLE__) && defined(__aarch64__) && \
+#if defined(__APPLE__) && WEBP_AARCH64 && \
     defined(__apple_build_version__) && (__apple_build_version__< 6020037)
 #define USE_VTBLQ
 #endif

src/dsp/lossless_neon.c

@@ -498,7 +498,7 @@ static void PredictorAdd13_NEON(const uint32_t* in, const uint32_t* upper,
 
 // vtbl?_u8 are marked unavailable for iOS arm64 with Xcode < 6.3, use
 // non-standard versions there.
-#if defined(__APPLE__) && defined(__aarch64__) && \
+#if defined(__APPLE__) && WEBP_AARCH64 && \
     defined(__apple_build_version__) && (__apple_build_version__< 6020037)
 #define USE_VTBLQ
 #endif

src/dsp/neon.h

@@ -21,7 +21,7 @@
 // Right now, some intrinsics functions seem slower, so we disable them
 // everywhere except newer clang/gcc or aarch64 where the inline assembly is
 // incompatible.
-#if LOCAL_CLANG_PREREQ(3,8) || LOCAL_GCC_PREREQ(4,9) || defined(__aarch64__)
+#if LOCAL_CLANG_PREREQ(3,8) || LOCAL_GCC_PREREQ(4,9) || WEBP_AARCH64
 #define WEBP_USE_INTRINSICS   // use intrinsics when possible
 #endif
 
@@ -46,7 +46,7 @@
 // if using intrinsics, this flag avoids some functions that make gcc-4.6.3
 // crash ("internal compiler error: in immed_double_const, at emit-rtl.").
 // (probably similar to gcc.gnu.org/bugzilla/show_bug.cgi?id=48183)
-#if !(LOCAL_CLANG_PREREQ(3,8) || LOCAL_GCC_PREREQ(4,8) || defined(__aarch64__))
+#if !(LOCAL_CLANG_PREREQ(3,8) || LOCAL_GCC_PREREQ(4,8) || WEBP_AARCH64)
 #define WORK_AROUND_GCC
 #endif
 

src/dsp/quant.h

@@ -22,7 +22,7 @@
 #define IsFlat IsFlat_NEON
 
 static uint32_t horizontal_add_uint32x4(const uint32x4_t a) {
-#if defined(__aarch64__)
+#if WEBP_AARCH64
   return vaddvq_u32(a);
 #else
   const uint64x2_t b = vpaddlq_u32(a);

src/utils/bit_reader_utils.c

@@ -15,6 +15,7 @@
 #include "src/webp/config.h"
 #endif
 
+#include "src/dsp/cpu.h"
 #include "src/utils/bit_reader_inl_utils.h"
 #include "src/utils/utils.h"
 
@@ -121,7 +122,7 @@ int32_t VP8GetSignedValue(VP8BitReader* const br, int bits,
 
 #define VP8L_LOG8_WBITS 4  // Number of bytes needed to store VP8L_WBITS bits.
 
-#if defined(__arm__) || defined(_M_ARM) || defined(__aarch64__) || \
+#if defined(__arm__) || defined(_M_ARM) || WEBP_AARCH64 || \
     defined(__i386__) || defined(_M_IX86) || \
     defined(__x86_64__) || defined(_M_X64)
 #define VP8L_USE_FAST_LOAD

src/utils/bit_reader_utils.h

@@ -19,6 +19,7 @@
 #ifdef _MSC_VER
 #include <stdlib.h>  // _byteswap_ulong
 #endif
+#include "src/dsp/cpu.h"
 #include "src/webp/types.h"
 
 // Warning! This macro triggers quite some MACRO wizardry around func signature!
@@ -64,7 +65,7 @@ extern "C" {
 #define BITS 56
 #elif defined(__arm__) || defined(_M_ARM)      // ARM
 #define BITS 24
-#elif defined(__aarch64__)                     // ARM 64bit
+#elif WEBP_AARCH64                             // ARM 64bit
 #define BITS 56
 #elif defined(__mips__)                        // MIPS
 #define BITS 24