gfx/skia/trunk/src/opts/SkUtils_opts_SSE2.cpp

changeset 0
6474c204b198
     1.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     1.2 +++ b/gfx/skia/trunk/src/opts/SkUtils_opts_SSE2.cpp	Wed Dec 31 06:09:35 2014 +0100
     1.3 @@ -0,0 +1,71 @@
     1.4 +
     1.5 +/*
     1.6 + * Copyright 2009 The Android Open Source Project
     1.7 + *
     1.8 + * Use of this source code is governed by a BSD-style license that can be
     1.9 + * found in the LICENSE file.
    1.10 + */
    1.11 +
    1.12 +
    1.13 +#include <emmintrin.h>
    1.14 +#include "SkUtils_opts_SSE2.h"
    1.15 +
    1.16 +void sk_memset16_SSE2(uint16_t *dst, uint16_t value, int count)
    1.17 +{
    1.18 +    SkASSERT(dst != NULL && count >= 0);
    1.19 +
    1.20 +    // dst must be 2-byte aligned.
    1.21 +    SkASSERT((((size_t) dst) & 0x01) == 0);
    1.22 +
    1.23 +    if (count >= 32) {
    1.24 +        while (((size_t)dst) & 0x0F) {
    1.25 +            *dst++ = value;
    1.26 +            --count;
    1.27 +        }
    1.28 +        __m128i *d = reinterpret_cast<__m128i*>(dst);
    1.29 +        __m128i value_wide = _mm_set1_epi16(value);
    1.30 +        while (count >= 32) {
    1.31 +            _mm_store_si128(d    , value_wide);
    1.32 +            _mm_store_si128(d + 1, value_wide);
    1.33 +            _mm_store_si128(d + 2, value_wide);
    1.34 +            _mm_store_si128(d + 3, value_wide);
    1.35 +            d += 4;
    1.36 +            count -= 32;
    1.37 +        }
    1.38 +        dst = reinterpret_cast<uint16_t*>(d);
    1.39 +    }
    1.40 +    while (count > 0) {
    1.41 +        *dst++ = value;
    1.42 +        --count;
    1.43 +    }
    1.44 +}
    1.45 +
    1.46 +void sk_memset32_SSE2(uint32_t *dst, uint32_t value, int count)
    1.47 +{
    1.48 +    SkASSERT(dst != NULL && count >= 0);
    1.49 +
    1.50 +    // dst must be 4-byte aligned.
    1.51 +    SkASSERT((((size_t) dst) & 0x03) == 0);
    1.52 +
    1.53 +    if (count >= 16) {
    1.54 +        while (((size_t)dst) & 0x0F) {
    1.55 +            *dst++ = value;
    1.56 +            --count;
    1.57 +        }
    1.58 +        __m128i *d = reinterpret_cast<__m128i*>(dst);
    1.59 +        __m128i value_wide = _mm_set1_epi32(value);
    1.60 +        while (count >= 16) {
    1.61 +            _mm_store_si128(d    , value_wide);
    1.62 +            _mm_store_si128(d + 1, value_wide);
    1.63 +            _mm_store_si128(d + 2, value_wide);
    1.64 +            _mm_store_si128(d + 3, value_wide);
    1.65 +            d += 4;
    1.66 +            count -= 16;
    1.67 +        }
    1.68 +        dst = reinterpret_cast<uint32_t*>(d);
    1.69 +    }
    1.70 +    while (count > 0) {
    1.71 +        *dst++ = value;
    1.72 +        --count;
    1.73 +    }
    1.74 +}

mercurial