From 8c38010f7dcefe57f8370bb106a625007e514567 Mon Sep 17 00:00:00 2001 From: Alexander Date: Sun, 10 Sep 2017 03:04:34 +0300 Subject: [PATCH] Speedup other 2L convertions --- src/libImaging/Convert.c | 37 +++++++++++++++++++++++++++++++++---- 1 file changed, 33 insertions(+), 4 deletions(-) diff --git a/src/libImaging/Convert.c b/src/libImaging/Convert.c index 0295e1007..74d0d4a74 100644 --- a/src/libImaging/Convert.c +++ b/src/libImaging/Convert.c @@ -240,8 +240,24 @@ rgb2l(UINT8* out, const UINT8* in, int xsize) static void rgb2la(UINT8* out, const UINT8* in, int xsize) { - int x; - for (x = 0; x < xsize; x++, in += 4, out += 4) { + int x = 0; + __m128i coeff = _mm_set_epi16( + 0, 3735, 19235, 9798, 0, 3735, 19235, 9798); + for (; x < xsize - 3; x += 4, in += 16, out += 16) { + __m128i pix0, pix1; + __m128i source = _mm_loadu_si128((__m128i*)in); + pix0 = _mm_unpacklo_epi8(source, _mm_setzero_si128()); + pix1 = _mm_unpackhi_epi8(source, _mm_setzero_si128()); + pix0 = _mm_madd_epi16(pix0, coeff); + pix1 = _mm_madd_epi16(pix1, coeff); + pix0 = _mm_hadd_epi32(pix0, pix1); + pix0 = _mm_srli_epi32(pix0, 15); + pix0 = _mm_shuffle_epi8(pix0, _mm_set_epi8( + -1,12,12,12, -1,8,8,8, -1,4,4,4, -1,0,0,0)); + pix0 = _mm_or_si128(pix0, _mm_set1_epi32(0xff000000)); + _mm_storeu_si128((__m128i*)out, pix0); + } + for (; x < xsize; x++, in += 4, out += 4) { /* ITU-R Recommendation 601-2 (assuming nonlinear RGB) */ out[0] = out[1] = out[2] = L24(in) >> 16; out[3] = 255; @@ -251,9 +267,22 @@ rgb2la(UINT8* out, const UINT8* in, int xsize) static void rgb2i(UINT8* out_, const UINT8* in, int xsize) { - int x; + int x = 0; INT32* out = (INT32*) out_; - for (x = 0; x < xsize; x++, in += 4) + __m128i coeff = _mm_set_epi16( + 0, 3735, 19235, 9798, 0, 3735, 19235, 9798); + for (; x < xsize - 3; x += 4, in += 16, out += 4) { + __m128i pix0, pix1; + __m128i source = _mm_loadu_si128((__m128i*)in); + pix0 = _mm_unpacklo_epi8(source, _mm_setzero_si128()); + pix1 = _mm_unpackhi_epi8(source, _mm_setzero_si128()); + pix0 = _mm_madd_epi16(pix0, coeff); + pix1 = _mm_madd_epi16(pix1, coeff); + pix0 = _mm_hadd_epi32(pix0, pix1); + pix0 = _mm_srli_epi32(pix0, 15); + _mm_storeu_si128((__m128i*)out, pix0); + } + for (; x < xsize; x++, in += 4) *out++ = L24(in) >> 16; }