From c89358d6b93abbbb553f29445489ff853021ff18 Mon Sep 17 00:00:00 2001 From: Bartosz Taudul Date: Fri, 5 Jul 2019 18:17:50 +0200 Subject: [PATCH] Two ands less in SSE DXT1 compression. --- client/TracyDxt1.cpp | 21 ++++++--------------- 1 file changed, 6 insertions(+), 15 deletions(-) diff --git a/client/TracyDxt1.cpp b/client/TracyDxt1.cpp index e43a430e..74de820e 100644 --- a/client/TracyDxt1.cpp +++ b/client/TracyDxt1.cpp @@ -243,22 +243,13 @@ static tracy_force_inline uint64_t ProcessRGB( const uint8_t* src ) __m128i p0 = _mm_packus_epi16( m0, m1 ); - __m128i mask0 = _mm_set1_epi32( 0x00000003 ); - __m128i mask1 = _mm_set1_epi32( 0x00000300 ); - __m128i mask2 = _mm_set1_epi32( 0x00030000 ); - __m128i mask3 = _mm_set1_epi32( 0x03000000 ); + __m128i p1 = _mm_or_si128( _mm_srai_epi32( p0, 6 ), _mm_srai_epi32( p0, 12 ) ); + __m128i p2 = _mm_or_si128( _mm_srai_epi32( p0, 18 ), p0 ); + __m128i p3 = _mm_or_si128( p1, p2 ); + __m128i p4 = _mm_and_si128( p3, _mm_set1_epi32( 0xFF ) ); - __m128i p1 = _mm_and_si128( p0, mask0 ); - __m128i p2 = _mm_srai_epi32( _mm_and_si128( p0, mask1 ), 6 ); - __m128i p3 = _mm_srai_epi32( _mm_and_si128( p0, mask2 ), 12 ); - __m128i p4 = _mm_srai_epi32( _mm_and_si128( p0, mask3 ), 18 ); - - __m128i p5 = _mm_or_si128( p1, p2 ); - __m128i p6 = _mm_or_si128( p3, p4 ); - __m128i p7 = _mm_or_si128( p5, p6 ); - - __m128i p8 = _mm_packus_epi32( p7, p7 ); - __m128i p = _mm_packus_epi16( p8, p8 ); + __m128i p5 = _mm_packus_epi32( p4, p4 ); + __m128i p = _mm_packus_epi16( p5, p5 ); uint32_t vmin = _mm_cvtsi128_si32( min ); uint32_t vmax = _mm_cvtsi128_si32( max );