[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Openexr-devel] Vector-of-float to vector-of-half conversion?
From: |
Simon Eves |
Subject: |
Re: [Openexr-devel] Vector-of-float to vector-of-half conversion? |
Date: |
Thu, 21 Jul 2011 18:07:07 -0700 |
How about these...?
http://software.intel.com/sites/products/documentation/hpc/composerxe/en-us/cpp/mac/intref_cls/common/intref_bk_convertpost32_hf.htm
On Wed, Jul 20, 2011 at 4:59 PM, James Bowman <address@hidden> wrote:
> Here's my conversion code. It handles infinity, overflows and underflows.
> But it does not handle denormals or NaNs. It also truncates the mantissa,
> instead of rounding it. As I clock it, it's about 2X the speed of the
> conversion in "Half".
> #include <xmmintrin.h>
> #define Ki(x) _mm_set_epi32((x),(x),(x),(x))
> // Convert n floats at src to half-floats at dst. n is a multiple of 4
> static void qhalf(unsigned short *dst, const float *src, size_t n)
> {
> while (n) {
> __m128i full = _mm_loadu_si128((__m128i*)src);
> __m128i sign = full & Ki(1UL << 31); // extract
> sign
> __m128i mantissa = full & Ki((1 << 23) - 1); // extract
> mantissa
> __m128i exp8 = _mm_srli_epi32(full, 23) & Ki(0xff); // extract
> exponent
> __m128i is_huge = _mm_cmpgt_epi32(exp8, Ki(0x8f)); // means flush
> to INF
> __m128i is_tiny = _mm_cmplt_epi32(exp8, Ki(0x70)); // means flush
> to zero
> __m128i exp5 = _mm_add_epi32(exp8, Ki(-0x7f + 0xf)); // rebias
> exponent
> // INF is represented by exponent=0x1f and mantissa=0
> exp5 = _mm_and_si128(_mm_or_si128(exp5, is_huge), Ki(0x1f));
> mantissa = _mm_andnot_si128(is_huge, mantissa);
> __m128i half = _mm_slli_epi32(exp5, 10) | _mm_srli_epi32(mantissa, 13);
> // Underflow: force to zero
> half = _mm_andnot_si128(is_tiny, half);
> // apply sign
> half = _mm_or_si128(half, _mm_srai_epi32(sign, 16));
> _mm_storel_pi((__m64*)dst, (__m128)_mm_packs_epi32((__m128i)half,
> (__m128i)half));
> dst += 4;
> src += 4;
> n -= 4;
> }
> }
>
> On Tue, Jul 19, 2011 at 11:19 AM, Florian Kainz <address@hidden> wrote:
>>
>> Have any of you tried to convert vectors of 32-bit floating-point
>> numbers to vectors of 16-bit floating-point numbers in a way that
>> is faster than calling OpenEXR's half-from-float constructor for
>> each vector element? Maybe using MMX, SSE or AVX instructions?
>>
>> _______________________________________________
>> Openexr-devel mailing list
>> address@hidden
>> https://lists.nongnu.org/mailman/listinfo/openexr-devel
>
>
>
> --
> James Bowman
> http://www.excamera.com/sphinx/articles-openexr.html
>
>
>
> _______________________________________________
> Openexr-devel mailing list
> address@hidden
> https://lists.nongnu.org/mailman/listinfo/openexr-devel
>
>