c - 高效的 SSE FP `floor()`/`ceil()`/`round()` 没有 SSE4.1 的舍入函数？

Question

如何像这些函数一样将 float 的 __m128 vector 向上/向下舍入或舍入到最接近的整数？

• Round - roundf()
• Ceil - ceilf()或 SSE4.1 _mm_ceil_ps。
• Floor - floorf()或 SSE4.1 _mm_floor_ps。

我需要在没有 SSE4.1 roundps (_mm_floor_ps/_mm_ceil_ps/_mm_round_ps (x, _MM_FROUND_TO_NEAREST_INT |_MM_FROUND_NO_EXC)。roundps 也可以截断为零，但我不需要这个应用程序。

我可以使用 SSE3 和更早版本。 (没有 SSSE3 或 SSE4)

所以函数声明应该是这样的:

__m128 RoundSse( __m128 x )、__m128 CeilSse( __m128 x ) 和 __m128 FloorSse( __m128 x )。

Answer 1

我正在发布来自 http://dss.stephanierct.com/DevBlog/?p=8 的代码:

它应该采用按值形式(我只是从代码中删除了&，不确定是否可以):

static inline __m128 FloorSse(const __m128 x) {
    __m128i v0 = _mm_setzero_si128();
    __m128i v1 = _mm_cmpeq_epi32(v0, v0);
    __m128i ji = _mm_srli_epi32(v1, 25);
    __m128i tmp = _mm_slli_epi32(ji, 23); // I edited this (Added tmp) not sure about it
    __m128 j = _mm_castsi128_ps(tmp); //create vector 1.0f // I edited this not sure about it
    __m128i i = _mm_cvttps_epi32(x);
    __m128 fi = _mm_cvtepi32_ps(i);
    __m128 igx = _mm_cmpgt_ps(fi, x);
    j = _mm_and_ps(igx, j);
    return _mm_sub_ps(fi, j);
}

static inline __m128 CeilSse(const __m128 x) {
    __m128i v0 = _mm_setzero_si128();
    __m128i v1 = _mm_cmpeq_epi32(v0, v0);
    __m128i ji = _mm_srli_epi32(v1, 25);
    __m128i tmp = _mm_slli_epi32(ji, 23); // I edited this (Added tmp) not sure about it
    __m128 j = _mm_castsi128_ps(tmp); //create vector 1.0f // I edited this not sure about it
    __m128i i = _mm_cvttps_epi32(x);
    __m128 fi = _mm_cvtepi32_ps(i);
    __m128 igx = _mm_cmplt_ps(fi, x);
    j = _mm_and_ps(igx, j);
    return _mm_add_ps(fi, j);
}

static inline __m128 RoundSse(const __m128 a) {
    __m128 v0 = _mm_setzero_ps();             //generate the highest value < 2
    __m128 v1 = _mm_cmpeq_ps(v0, v0);
    __m128i tmp = _mm_castps_si128(v1); // I edited this (Added tmp) not sure about it
    tmp = _mm_srli_epi32(tmp, 2); // I edited this (Added tmp) not sure about it
    __m128 vNearest2 = _mm_castsi128_ps(tmp); // I edited this (Added tmp) not sure about it
    __m128i i = _mm_cvttps_epi32(a);
    __m128 aTrunc = _mm_cvtepi32_ps(i);        // truncate a
    __m128 rmd = _mm_sub_ps(a, aTrunc);        // get remainder
    __m128 rmd2 = _mm_mul_ps(rmd, vNearest2); // mul remainder by near 2 will yield the needed offset
    __m128i rmd2i = _mm_cvttps_epi32(rmd2);    // after being truncated of course
    __m128 rmd2Trunc = _mm_cvtepi32_ps(rmd2i);
    __m128 r = _mm_add_ps(aTrunc, rmd2Trunc);
    return r;
}


inline __m128 ModSee(const __m128 a, const __m128 aDiv) {
    __m128 c = _mm_div_ps(a, aDiv);
    __m128i i = _mm_cvttps_epi32(c);
    __m128 cTrunc = _mm_cvtepi32_ps(i);
    __m128 base = _mm_mul_ps(cTrunc, aDiv);
    __m128 r = _mm_sub_ps(a, base);
    return r;
}