ShuriZma
/
pcsx2
mirror of https://github.com/PCSX2/pcsx2.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

pcsx2: Add IPU dither block SSE2 implementation

This commit is contained in:
Jonathan Li 2019-10-07 17:50:06 +01:00
parent f9b51e2656
commit c533a65764
1 changed files with 65 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

65
pcsx2/IPU/IPUdither.cpp
View File

@ -21,7 +21,15 @@
#include "yuv2rgb.h" #include "yuv2rgb.h"
#include "mpeg2lib/Mpeg.h" #include "mpeg2lib/Mpeg.h"
void ipu_dither_reference(const macroblock_rgb32 &rgb32, macroblock_rgb16 &rgb16, int dte);
void ipu_dither_sse2(const macroblock_rgb32 &rgb32, macroblock_rgb16 &rgb16, int dte);
__ri void ipu_dither(const macroblock_rgb32 &rgb32, macroblock_rgb16 &rgb16, int dte) __ri void ipu_dither(const macroblock_rgb32 &rgb32, macroblock_rgb16 &rgb16, int dte)
{
ipu_dither_sse2(rgb32, rgb16, dte);
}
__ri void ipu_dither_reference(const macroblock_rgb32 &rgb32, macroblock_rgb16 &rgb16, int dte)
{ {
if (dte) { if (dte) {
// I'm guessing values are rounded down when clamping. // I'm guessing values are rounded down when clamping.
@ -55,3 +63,60 @@ __ri void ipu_dither(const macroblock_rgb32 &rgb32, macroblock_rgb16 &rgb16, int
} }
} }
} }
__ri void ipu_dither_sse2(const macroblock_rgb32 &rgb32, macroblock_rgb16 &rgb16, int dte)
{
const __m128i alpha_test = _mm_set1_epi16(0x40);
const __m128i dither_add_matrix[] = {
_mm_setr_epi32(0x00000000, 0x00000000, 0x00000000, 0x00010101),
_mm_setr_epi32(0x00020202, 0x00000000, 0x00030303, 0x00000000),
_mm_setr_epi32(0x00000000, 0x00010101, 0x00000000, 0x00000000),
_mm_setr_epi32(0x00030303, 0x00000000, 0x00020202, 0x00000000),
};
const __m128i dither_sub_matrix[] = {
_mm_setr_epi32(0x00040404, 0x00000000, 0x00030303, 0x00000000),
_mm_setr_epi32(0x00000000, 0x00020202, 0x00000000, 0x00010101),
_mm_setr_epi32(0x00030303, 0x00000000, 0x00040404, 0x00000000),
_mm_setr_epi32(0x00000000, 0x00010101, 0x00000000, 0x00020202),
};
for (int i = 0; i < 16; ++i) {
const __m128i dither_add = dither_add_matrix[i & 3];
const __m128i dither_sub = dither_sub_matrix[i & 3];
for (int n = 0; n < 2; ++n) {
__m128i rgba_8_0123 = _mm_load_si128(reinterpret_cast<const __m128i *>(&rgb32.c[i][n * 8]));
__m128i rgba_8_4567 = _mm_load_si128(reinterpret_cast<const __m128i *>(&rgb32.c[i][n * 8 + 4]));
// Dither and clamp
if (dte) {
rgba_8_0123 = _mm_adds_epu8(rgba_8_0123, dither_add);
rgba_8_0123 = _mm_subs_epu8(rgba_8_0123, dither_sub);
rgba_8_4567 = _mm_adds_epu8(rgba_8_4567, dither_add);
rgba_8_4567 = _mm_subs_epu8(rgba_8_4567, dither_sub);
}
// Split into channel components and extend to 16 bits
const __m128i rgba_16_0415 = _mm_unpacklo_epi8(rgba_8_0123, rgba_8_4567);
const __m128i rgba_16_2637 = _mm_unpackhi_epi8(rgba_8_0123, rgba_8_4567);
const __m128i rgba_32_0246 = _mm_unpacklo_epi8(rgba_16_0415, rgba_16_2637);
const __m128i rgba_32_1357 = _mm_unpackhi_epi8(rgba_16_0415, rgba_16_2637);
const __m128i rg_64_01234567 = _mm_unpacklo_epi8(rgba_32_0246, rgba_32_1357);
const __m128i ba_64_01234567 = _mm_unpackhi_epi8(rgba_32_0246, rgba_32_1357);
const __m128i zero = _mm_setzero_si128();
__m128i r = _mm_unpacklo_epi8(rg_64_01234567, zero);
__m128i g = _mm_unpackhi_epi8(rg_64_01234567, zero);
__m128i b = _mm_unpacklo_epi8(ba_64_01234567, zero);
__m128i a = _mm_unpackhi_epi8(ba_64_01234567, zero);
// Create RGBA
r = _mm_srli_epi16(r, 3);
g = _mm_slli_epi16(_mm_srli_epi16(g, 3), 5);
b = _mm_slli_epi16(_mm_srli_epi16(b, 3), 10);
a = _mm_slli_epi16(_mm_cmpeq_epi16(a, alpha_test), 15);
const __m128i rgba16 = _mm_or_si128(_mm_or_si128(r, g), _mm_or_si128(b, a));
_mm_store_si128(reinterpret_cast<__m128i *>(&rgb16.c[i][n * 8]), rgba16);
}
}
}