dany/libwebp
1
0
Fork 0
mirror of https://github.com/webmproject/libwebp.git synced 2025-10-31 18:35:41 +01:00
Code Issues Actions Packages Projects Releases Wiki Activity

wasm: Enable neon add,sub with sat builtins

Browse Source 
For performance testing

BUG=webp:352

Change-Id: I3fab48ee610437ac07dd603208972edf17c6f50b
This commit is contained in:
Scott LaVarnway
2017-07-14 12:49:13 -07:00
parent 09bcd9a397
commit 415b98ffad
1 changed files with 18 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
src/dsp/dec_wasm.c
View File

@@ -332,10 +332,18 @@ static void Transform(const int16_t* in, uint8_t* dst, int do_two) {
// For testing purposes. // For testing purposes.
// #define ENABLE_X86_BUILTIN_ADDSUB_SAT // #define ENABLE_X86_BUILTIN_ADDSUB_SAT
#if defined(__aarch64__)
// For now, this only works for aarch64
// TODO(slavarnway): Find a workaround for the ICE when using clang-3.9.1
#define ENABLE_NEON_BUILTIN_ADDSUB_SAT
#endif
static WEBP_INLINE uint8x16 uint8x16_add_sat(const uint8x16 a, static WEBP_INLINE uint8x16 uint8x16_add_sat(const uint8x16 a,
const uint8x16 b) { const uint8x16 b) {
#ifdef ENABLE_X86_BUILTIN_ADDSUB_SAT #if defined(ENABLE_X86_BUILTIN_ADDSUB_SAT)
return (uint8x16)__builtin_ia32_paddusb128(a, b); return (uint8x16)__builtin_ia32_paddusb128(a, b);
#elif defined(ENABLE_NEON_BUILTIN_ADDSUB_SAT)
return (uint8x16)__builtin_neon_vqaddq_v(a, b, 48);
#else #else
// Generic implementation for non-x86 // Generic implementation for non-x86
const uint8x16 zero = splat_uint8(0); const uint8x16 zero = splat_uint8(0);
@@ -352,8 +360,10 @@ static WEBP_INLINE uint8x16 uint8x16_add_sat(const uint8x16 a,
} }
static WEBP_INLINE int8x16 int8x16_add_sat(const int8x16 a, const int8x16 b) { static WEBP_INLINE int8x16 int8x16_add_sat(const int8x16 a, const int8x16 b) {
#ifdef ENABLE_X86_BUILTIN_ADDSUB_SAT #if defined(ENABLE_X86_BUILTIN_ADDSUB_SAT)
return (int8x16)__builtin_ia32_paddsb128(a, b); return (int8x16)__builtin_ia32_paddsb128(a, b);
#elif defined(ENABLE_NEON_BUILTIN_ADDSUB_SAT)
return (int8x16)__builtin_neon_vqaddq_v(a, b, 32);
#else #else
// Generic implementation for non-x86 // Generic implementation for non-x86
const int8x16 zero = splat_uint8(0); const int8x16 zero = splat_uint8(0);
@@ -372,8 +382,10 @@ static WEBP_INLINE int8x16 int8x16_add_sat(const int8x16 a, const int8x16 b) {
static WEBP_INLINE uint8x16 uint8x16_sub_sat(const uint8x16 a, static WEBP_INLINE uint8x16 uint8x16_sub_sat(const uint8x16 a,
const uint8x16 b) { const uint8x16 b) {
#ifdef ENABLE_X86_BUILTIN_ADDSUB_SAT #if defined(ENABLE_X86_BUILTIN_ADDSUB_SAT)
return (uint8x16)__builtin_ia32_psubusb128(a, b); return (uint8x16)__builtin_ia32_psubusb128(a, b);
#elif defined(ENABLE_NEON_BUILTIN_ADDSUB_SAT)
return (int8x16)__builtin_neon_vqsubq_v(a, b, 48);
#else #else
// Generic implementation for non-x86 // Generic implementation for non-x86
const uint8x16 zero = splat_uint8(0); const uint8x16 zero = splat_uint8(0);
@@ -390,8 +402,10 @@ static WEBP_INLINE uint8x16 uint8x16_sub_sat(const uint8x16 a,
} }
static WEBP_INLINE int8x16 int8x16_sub_sat(const int8x16 a, const int8x16 b) { static WEBP_INLINE int8x16 int8x16_sub_sat(const int8x16 a, const int8x16 b) {
#ifdef ENABLE_X86_BUILTIN_ADDSUB_SAT #if defined(ENABLE_X86_BUILTIN_ADDSUB_SAT)
return (int8x16)__builtin_ia32_psubsb128(a, b); return (int8x16)__builtin_ia32_psubsb128(a, b);
#elif defined(ENABLE_NEON_BUILTIN_ADDSUB_SAT)
return (int8x16)__builtin_neon_vqsubq_v(a, b, 32);
#else #else
// Generic implementation for non-x86 // Generic implementation for non-x86
const int8x16 zero = splat_uint8(0); const int8x16 zero = splat_uint8(0);