diff options
Diffstat (limited to 'src/core/SkPackBits.cpp')
-rw-r--r-- | src/core/SkPackBits.cpp | 407 |
1 files changed, 407 insertions, 0 deletions
diff --git a/src/core/SkPackBits.cpp b/src/core/SkPackBits.cpp new file mode 100644 index 0000000..3e92841 --- /dev/null +++ b/src/core/SkPackBits.cpp @@ -0,0 +1,407 @@ +#include "SkPackBits.h" + +#define GATHER_STATSx + +static inline void small_memcpy(void* SK_RESTRICT dst, + const void* SK_RESTRICT src, int n) { + SkASSERT(n > 0 && n <= 15); + uint8_t* d = (uint8_t*)dst; + const uint8_t* s = (const uint8_t*)src; + switch (n) { + case 15: *d++ = *s++; + case 14: *d++ = *s++; + case 13: *d++ = *s++; + case 12: *d++ = *s++; + case 11: *d++ = *s++; + case 10: *d++ = *s++; + case 9: *d++ = *s++; + case 8: *d++ = *s++; + case 7: *d++ = *s++; + case 6: *d++ = *s++; + case 5: *d++ = *s++; + case 4: *d++ = *s++; + case 3: *d++ = *s++; + case 2: *d++ = *s++; + case 1: *d++ = *s++; + case 0: break; + } +} + +static inline void small_memset(void* dst, uint8_t value, int n) { + SkASSERT(n > 0 && n <= 15); + uint8_t* d = (uint8_t*)dst; + switch (n) { + case 15: *d++ = value; + case 14: *d++ = value; + case 13: *d++ = value; + case 12: *d++ = value; + case 11: *d++ = value; + case 10: *d++ = value; + case 9: *d++ = value; + case 8: *d++ = value; + case 7: *d++ = value; + case 6: *d++ = value; + case 5: *d++ = value; + case 4: *d++ = value; + case 3: *d++ = value; + case 2: *d++ = value; + case 1: *d++ = value; + case 0: break; + } +} + +// can we do better for small counts with our own inlined memcpy/memset? + +#define PB_MEMSET(addr, value, count) \ +do { \ +if ((count) > 15) { \ +memset(addr, value, count); \ +} else { \ +small_memset(addr, value, count); \ +} \ +} while (0) + +#define PB_MEMCPY(dst, src, count) \ +do { \ + if ((count) > 15) { \ + memcpy(dst, src, count); \ + } else { \ + small_memcpy(dst, src, count); \ + } \ +} while (0) + +/////////////////////////////////////////////////////////////////////////////// + +#ifdef GATHER_STATS + static int gMemSetBuckets[129]; + static int gMemCpyBuckets[129]; + static int gCounter; + +static void register_memset_count(int n) { + SkASSERT((unsigned)n <= 128); + gMemSetBuckets[n] += 1; + gCounter += 1; + + if ((gCounter & 0xFF) == 0) { + SkDebugf("----- packbits memset stats: "); + for (size_t i = 0; i < SK_ARRAY_COUNT(gMemSetBuckets); i++) { + if (gMemSetBuckets[i]) { + SkDebugf(" %d:%d", i, gMemSetBuckets[i]); + } + } + } +} +static void register_memcpy_count(int n) { + SkASSERT((unsigned)n <= 128); + gMemCpyBuckets[n] += 1; + gCounter += 1; + + if ((gCounter & 0x1FF) == 0) { + SkDebugf("----- packbits memcpy stats: "); + for (size_t i = 0; i < SK_ARRAY_COUNT(gMemCpyBuckets); i++) { + if (gMemCpyBuckets[i]) { + SkDebugf(" %d:%d", i, gMemCpyBuckets[i]); + } + } + } +} +#else +#define register_memset_count(n) +#define register_memcpy_count(n) +#endif + + +/////////////////////////////////////////////////////////////////////////////// + +size_t SkPackBits::ComputeMaxSize16(int count) { + // worst case is the number of 16bit values (times 2) + + // 1 byte per (up to) 128 entries. + return ((count + 127) >> 7) + (count << 1); +} + +size_t SkPackBits::ComputeMaxSize8(int count) { + // worst case is the number of 8bit values + 1 byte per (up to) 128 entries. + return ((count + 127) >> 7) + count; +} + +static uint8_t* flush_same16(uint8_t dst[], uint16_t value, int count) { + while (count > 0) { + int n = count; + if (n > 128) { + n = 128; + } + *dst++ = (uint8_t)(n - 1); + *dst++ = (uint8_t)(value >> 8); + *dst++ = (uint8_t)value; + count -= n; + } + return dst; +} + +static uint8_t* flush_same8(uint8_t dst[], uint8_t value, int count) { + while (count > 0) { + int n = count; + if (n > 128) { + n = 128; + } + *dst++ = (uint8_t)(n - 1); + *dst++ = (uint8_t)value; + count -= n; + } + return dst; +} + +static uint8_t* flush_diff16(uint8_t SK_RESTRICT dst[], + const uint16_t SK_RESTRICT src[], int count) { + while (count > 0) { + int n = count; + if (n > 128) { + n = 128; + } + *dst++ = (uint8_t)(n + 127); + PB_MEMCPY(dst, src, n * sizeof(uint16_t)); + src += n; + dst += n * sizeof(uint16_t); + count -= n; + } + return dst; +} + +static uint8_t* flush_diff8(uint8_t SK_RESTRICT dst[], + const uint8_t SK_RESTRICT src[], int count) { + while (count > 0) { + int n = count; + if (n > 128) { + n = 128; + } + *dst++ = (uint8_t)(n + 127); + PB_MEMCPY(dst, src, n); + src += n; + dst += n; + count -= n; + } + return dst; +} + +size_t SkPackBits::Pack16(const uint16_t SK_RESTRICT src[], int count, + uint8_t SK_RESTRICT dst[]) { + uint8_t* origDst = dst; + const uint16_t* stop = src + count; + + for (;;) { + count = stop - src; + SkASSERT(count >= 0); + if (count == 0) { + return dst - origDst; + } + if (1 == count) { + *dst++ = 0; + *dst++ = (uint8_t)(*src >> 8); + *dst++ = (uint8_t)*src; + return dst - origDst; + } + + unsigned value = *src; + const uint16_t* s = src + 1; + + if (*s == value) { // accumulate same values... + do { + s++; + if (s == stop) { + break; + } + } while (*s == value); + dst = flush_same16(dst, value, s - src); + } else { // accumulate diff values... + do { + if (++s == stop) { + goto FLUSH_DIFF; + } + } while (*s != s[-1]); + s -= 1; // back up so we don't grab one of the "same" values that follow + FLUSH_DIFF: + dst = flush_diff16(dst, src, s - src); + } + src = s; + } +} + +size_t SkPackBits::Pack8(const uint8_t SK_RESTRICT src[], int count, + uint8_t SK_RESTRICT dst[]) { + uint8_t* origDst = dst; + const uint8_t* stop = src + count; + + for (;;) { + count = stop - src; + SkASSERT(count >= 0); + if (count == 0) { + return dst - origDst; + } + if (1 == count) { + *dst++ = 0; + *dst++ = *src; + return dst - origDst; + } + + unsigned value = *src; + const uint8_t* s = src + 1; + + if (*s == value) { // accumulate same values... + do { + s++; + if (s == stop) { + break; + } + } while (*s == value); + dst = flush_same8(dst, value, s - src); + } else { // accumulate diff values... + do { + if (++s == stop) { + goto FLUSH_DIFF; + } + // only stop if we hit 3 in a row, + // otherwise we get bigger than compuatemax + } while (*s != s[-1] || s[-1] != s[-2]); + s -= 2; // back up so we don't grab the "same" values that follow + FLUSH_DIFF: + dst = flush_diff8(dst, src, s - src); + } + src = s; + } +} + +#include "SkUtils.h" + +int SkPackBits::Unpack16(const uint8_t SK_RESTRICT src[], size_t srcSize, + uint16_t SK_RESTRICT dst[]) { + uint16_t* origDst = dst; + const uint8_t* stop = src + srcSize; + + while (src < stop) { + unsigned n = *src++; + if (n <= 127) { // repeat count (n + 1) + n += 1; + sk_memset16(dst, (src[0] << 8) | src[1], n); + src += 2; + } else { // same count (n - 127) + n -= 127; + PB_MEMCPY(dst, src, n * sizeof(uint16_t)); + src += n * sizeof(uint16_t); + } + dst += n; + } + SkASSERT(src == stop); + return dst - origDst; +} + +int SkPackBits::Unpack8(const uint8_t SK_RESTRICT src[], size_t srcSize, + uint8_t SK_RESTRICT dst[]) { + uint8_t* origDst = dst; + const uint8_t* stop = src + srcSize; + + while (src < stop) { + unsigned n = *src++; + if (n <= 127) { // repeat count (n + 1) + n += 1; + PB_MEMSET(dst, *src++, n); + } else { // same count (n - 127) + n -= 127; + PB_MEMCPY(dst, src, n); + src += n; + } + dst += n; + } + SkASSERT(src == stop); + return dst - origDst; +} + +enum UnpackState { + CLEAN_STATE, + REPEAT_BYTE_STATE, + COPY_SRC_STATE +}; + +void SkPackBits::Unpack8(uint8_t SK_RESTRICT dst[], size_t dstSkip, + size_t dstWrite, const uint8_t SK_RESTRICT src[]) { + if (dstWrite == 0) { + return; + } + + UnpackState state = CLEAN_STATE; + size_t stateCount = 0; + + // state 1: do the skip-loop + while (dstSkip > 0) { + unsigned n = *src++; + if (n <= 127) { // repeat count (n + 1) + n += 1; + if (n > dstSkip) { + state = REPEAT_BYTE_STATE; + stateCount = n - dstSkip; + n = dstSkip; + // we don't increment src here, since its needed in stage 2 + } else { + src++; // skip the src byte + } + } else { // same count (n - 127) + n -= 127; + if (n > dstSkip) { + state = COPY_SRC_STATE; + stateCount = n - dstSkip; + n = dstSkip; + } + src += n; + } + dstSkip -= n; + } + + // stage 2: perform any catchup from the skip-stage + if (stateCount > dstWrite) { + stateCount = dstWrite; + } + switch (state) { + case REPEAT_BYTE_STATE: + SkASSERT(stateCount > 0); + register_memset_count(stateCount); + PB_MEMSET(dst, *src++, stateCount); + break; + case COPY_SRC_STATE: + SkASSERT(stateCount > 0); + register_memcpy_count(stateCount); + PB_MEMCPY(dst, src, stateCount); + src += stateCount; + break; + default: + SkASSERT(stateCount == 0); + break; + } + dst += stateCount; + dstWrite -= stateCount; + + // copy at most dstWrite bytes into dst[] + while (dstWrite > 0) { + unsigned n = *src++; + if (n <= 127) { // repeat count (n + 1) + n += 1; + if (n > dstWrite) { + n = dstWrite; + } + register_memset_count(n); + PB_MEMSET(dst, *src++, n); + } else { // same count (n - 127) + n -= 127; + if (n > dstWrite) { + n = dstWrite; + } + register_memcpy_count(n); + PB_MEMCPY(dst, src, n); + src += n; + } + dst += n; + dstWrite -= n; + } + SkASSERT(0 == dstWrite); +} + + + |