introduce WEBP_REFERENCE_IMPLEMENTATION compile option

This flag will make the code use no uint64, no asm, and no fancy
trick, but instead aim at being as simple and straightforward as
possible.
Main use is to help emscripten generate proper JS code.
More code needs to be simplified later.

Also: tune the BITS values to be 24 and make use of WEBP_RIGHT_JUSTIFY
Here are the typical timing for decoding a large image:

        ARM7-a:
        dwebp_justify_32_neon Time to decode picture: 3.280s
        dwebp_justify_24_neon Time to decode picture: 2.640s
        dwebp_justify_16_neon Time to decode picture: 2.723s
        dwebp_justify_8_neon Time to decode picture: 2.802s
        dwebp_justify_32 Time to decode picture: 4.264s
        dwebp_justify_24 Time to decode picture: 3.696s
        dwebp_justify_16 Time to decode picture: 3.779s
        dwebp_justify_8 Time to decode picture: 3.834s
        dwebp_32_neon Time to decode picture: 4.010s
        dwebp_24_neon Time to decode picture: 2.725s
        dwebp_16_neon Time to decode picture: 2.852s
        dwebp_8_neon Time to decode picture: 2.778s
        dwebp_32 Time to decode picture: 4.587s
        dwebp_24 Time to decode picture: 3.800s
        dwebp_16 Time to decode picture: 3.902s
        dwebp_8 Time to decode picture: 3.815s
        REFERENCE (HEAD) Time to decode picture: 3.818s

        x86_64:
        dwebp_justify_32 Time to decode picture: 0.473s
        dwebp_justify_24 Time to decode picture: 0.434s
        dwebp_justify_16 Time to decode picture: 0.450s
        dwebp_justify_8 Time to decode picture: 0.467s
        dwebp_32 Time to decode picture: 0.474s
        dwebp_24 Time to decode picture: 0.468s
        dwebp_16 Time to decode picture: 0.468s
        dwebp_8 Time to decode picture: 0.481s
        REFERENCE (HEAD) Time to decode picture: 0.436s

        i386:
        dwebp_justify_32 Time to decode picture: 0.723s
        dwebp_justify_24 Time to decode picture: 0.618s
        dwebp_justify_16 Time to decode picture: 0.626s
        dwebp_justify_8 Time to decode picture: 0.651s
        dwebp_32 Time to decode picture: 0.744s
        dwebp_24 Time to decode picture: 0.627s
        dwebp_16 Time to decode picture: 0.642s
        dwebp_8 Time to decode picture: 0.642s

Change-Id: Ie56c7235733a24f94fbfc2e4351aae36ec39c225
This commit is contained in:
skal 2013-02-14 15:46:12 +01:00
parent 3383885799
commit b7490f8553
3 changed files with 37 additions and 16 deletions

View File

@ -426,11 +426,16 @@ static void HE8uv(uint8_t *dst) { // horizontal
}
// helper for chroma-DC predictions
static WEBP_INLINE void Put8x8uv(uint64_t v, uint8_t* dst) {
static WEBP_INLINE void Put8x8uv(uint8_t value, uint8_t* dst) {
int j;
#ifndef WEBP_REFERENCE_IMPLEMENTATION
const uint64_t v = (uint64_t)value * 0x0101010101010101ULL;
for (j = 0; j < 8; ++j) {
*(uint64_t*)(dst + j * BPS) = v;
}
#else
for (j = 0; j < 8; ++j) memset(dst + j * BPS, value, 8);
#endif
}
static void DC8uv(uint8_t *dst) { // DC
@ -439,7 +444,7 @@ static void DC8uv(uint8_t *dst) { // DC
for (i = 0; i < 8; ++i) {
dc0 += dst[i - BPS] + dst[-1 + i * BPS];
}
Put8x8uv((uint64_t)((dc0 >> 4) * 0x0101010101010101ULL), dst);
Put8x8uv(dc0 >> 4, dst);
}
static void DC8uvNoLeft(uint8_t *dst) { // DC with no left samples
@ -448,7 +453,7 @@ static void DC8uvNoLeft(uint8_t *dst) { // DC with no left samples
for (i = 0; i < 8; ++i) {
dc0 += dst[i - BPS];
}
Put8x8uv((uint64_t)((dc0 >> 3) * 0x0101010101010101ULL), dst);
Put8x8uv(dc0 >> 3, dst);
}
static void DC8uvNoTop(uint8_t *dst) { // DC with no top samples
@ -457,11 +462,11 @@ static void DC8uvNoTop(uint8_t *dst) { // DC with no top samples
for (i = 0; i < 8; ++i) {
dc0 += dst[-1 + i * BPS];
}
Put8x8uv((uint64_t)((dc0 >> 3) * 0x0101010101010101ULL), dst);
Put8x8uv(dc0 >> 3, dst);
}
static void DC8uvNoTopLeft(uint8_t *dst) { // DC with nothing
Put8x8uv(0x8080808080808080ULL, dst);
Put8x8uv(0x80, dst);
}
//------------------------------------------------------------------------------

View File

@ -1079,20 +1079,27 @@ static void CopyOrSwap(const uint32_t* src, int num_pixels, uint8_t* dst,
const uint32_t* const src_end = src + num_pixels;
while (src < src_end) {
uint32_t argb = *src++;
#if !defined(WEBP_REFERENCE_IMPLEMENTATION)
#if !defined(__BIG_ENDIAN__) && (defined(__i386__) || defined(__x86_64__))
__asm__ volatile("bswap %0" : "=r"(argb) : "0"(argb));
*(uint32_t*)dst = argb;
dst += sizeof(argb);
#elif !defined(__BIG_ENDIAN__) && defined(_MSC_VER)
argb = _byteswap_ulong(argb);
*(uint32_t*)dst = argb;
dst += sizeof(argb);
#else
*dst++ = (argb >> 24) & 0xff;
*dst++ = (argb >> 16) & 0xff;
*dst++ = (argb >> 8) & 0xff;
*dst++ = (argb >> 0) & 0xff;
dst[0] = (argb >> 24) & 0xff;
dst[1] = (argb >> 16) & 0xff;
dst[2] = (argb >> 8) & 0xff;
dst[3] = (argb >> 0) & 0xff;
#endif
#else // WEBP_REFERENCE_IMPLEMENTATION
dst[0] = (argb >> 24) & 0xff;
dst[1] = (argb >> 16) & 0xff;
dst[2] = (argb >> 8) & 0xff;
dst[3] = (argb >> 0) & 0xff;
#endif
dst += sizeof(argb);
}
} else {
memcpy(dst, src, num_pixels * sizeof(*src));

View File

@ -56,20 +56,29 @@ extern "C" {
// -> we're back to height active 'value_' bits (marked 'v') and BITS cached
// bits (marked 'B')
//
// The right-justify strategy tends to use less shifts, so let's use it:
#define USE_RIGHT_JUSTIFY
// The right-justify strategy tends to use less shifts and is often faster.
//------------------------------------------------------------------------------
// BITS can be either 32, 24, 16 or 8.
// Pick values that fit natural register size.
#if !defined(WEBP_REFERENCE_IMPLEMENTATION)
#define USE_RIGHT_JUSTIFY
#if defined(__i386__) || defined(_M_IX86) // x86 32bit
#define BITS 16
#elif defined(__arm__) || defined(_M_ARM) // ARM
#define BITS 8
#define BITS 24
#else // reasonable default
#define BITS 32
#define BITS 24
#endif
#else // reference choices
#define USE_RIGHT_JUSTIFY
#define BITS 8
#endif
//------------------------------------------------------------------------------