zdoom/gzdoom
0
0
Fork 0
mirror of https://github.com/ZDoom/gzdoom.git synced 2025-02-21 11:31:41 +00:00
Code Issues Projects Releases Packages Wiki Activity

Fix typo where __m128 should have been __m128i

Browse source
This commit is contained in:
Magnus Norddahl 2017-02-17 23:03:48 +01:00
parent 4240a15f3a
commit 89a2e434d4
2 changed files with 65 additions and 65 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

120
src/swrenderer/drawers/r_draw_wall32.h
View file

@ -89,7 +89,7 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8);
// Blend // Blend
__m128 outcolor = fgcolor; __m128i outcolor = fgcolor;
outcolor = _mm_packus_epi16(outcolor, _mm_setzero_si128()); outcolor = _mm_packus_epi16(outcolor, _mm_setzero_si128());
dest[offset] = _mm_cvtsi128_si32(outcolor); dest[offset] = _mm_cvtsi128_si32(outcolor);
@ -157,7 +157,7 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8);
// Blend // Blend
__m128 outcolor = fgcolor; __m128i outcolor = fgcolor;
outcolor = _mm_packus_epi16(outcolor, _mm_setzero_si128()); outcolor = _mm_packus_epi16(outcolor, _mm_setzero_si128());
dest[offset] = _mm_cvtsi128_si32(outcolor); dest[offset] = _mm_cvtsi128_si32(outcolor);
@ -225,7 +225,7 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8);
// Blend // Blend
__m128 outcolor = fgcolor; __m128i outcolor = fgcolor;
outcolor = _mm_packus_epi16(outcolor, _mm_setzero_si128()); outcolor = _mm_packus_epi16(outcolor, _mm_setzero_si128());
dest[offset] = _mm_cvtsi128_si32(outcolor); dest[offset] = _mm_cvtsi128_si32(outcolor);
@ -306,7 +306,7 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8);
// Blend // Blend
__m128 outcolor = fgcolor; __m128i outcolor = fgcolor;
outcolor = _mm_packus_epi16(outcolor, _mm_setzero_si128()); outcolor = _mm_packus_epi16(outcolor, _mm_setzero_si128());
dest[offset] = _mm_cvtsi128_si32(outcolor); dest[offset] = _mm_cvtsi128_si32(outcolor);
@ -377,10 +377,10 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8);
// Blend // Blend
__m128 alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); __m128i alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256 alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256
__m128 inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha); __m128i inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha);
fgcolor = _mm_mullo_epi16(fgcolor, alpha); fgcolor = _mm_mullo_epi16(fgcolor, alpha);
bgcolor = _mm_mullo_epi16(bgcolor, inv_alpha); bgcolor = _mm_mullo_epi16(bgcolor, inv_alpha);
__m128i outcolor = _mm_srli_epi16(_mm_add_epi16(fgcolor, bgcolor), 8); __m128i outcolor = _mm_srli_epi16(_mm_add_epi16(fgcolor, bgcolor), 8);
@ -452,10 +452,10 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8);
// Blend // Blend
__m128 alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); __m128i alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256 alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256
__m128 inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha); __m128i inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha);
fgcolor = _mm_mullo_epi16(fgcolor, alpha); fgcolor = _mm_mullo_epi16(fgcolor, alpha);
bgcolor = _mm_mullo_epi16(bgcolor, inv_alpha); bgcolor = _mm_mullo_epi16(bgcolor, inv_alpha);
__m128i outcolor = _mm_srli_epi16(_mm_add_epi16(fgcolor, bgcolor), 8); __m128i outcolor = _mm_srli_epi16(_mm_add_epi16(fgcolor, bgcolor), 8);
@ -527,10 +527,10 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8);
// Blend // Blend
__m128 alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); __m128i alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256 alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256
__m128 inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha); __m128i inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha);
fgcolor = _mm_mullo_epi16(fgcolor, alpha); fgcolor = _mm_mullo_epi16(fgcolor, alpha);
bgcolor = _mm_mullo_epi16(bgcolor, inv_alpha); bgcolor = _mm_mullo_epi16(bgcolor, inv_alpha);
__m128i outcolor = _mm_srli_epi16(_mm_add_epi16(fgcolor, bgcolor), 8); __m128i outcolor = _mm_srli_epi16(_mm_add_epi16(fgcolor, bgcolor), 8);
@ -615,10 +615,10 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8);
// Blend // Blend
__m128 alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); __m128i alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256 alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256
__m128 inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha); __m128i inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha);
fgcolor = _mm_mullo_epi16(fgcolor, alpha); fgcolor = _mm_mullo_epi16(fgcolor, alpha);
bgcolor = _mm_mullo_epi16(bgcolor, inv_alpha); bgcolor = _mm_mullo_epi16(bgcolor, inv_alpha);
__m128i outcolor = _mm_srli_epi16(_mm_add_epi16(fgcolor, bgcolor), 8); __m128i outcolor = _mm_srli_epi16(_mm_add_epi16(fgcolor, bgcolor), 8);
@ -693,15 +693,15 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8);
// Blend // Blend
__m128 alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); __m128i alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256 alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256
__m128 inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha); __m128i inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha);
__m128 bgalpha = _mm_mullo_epi16(destalpha, alpha); __m128i bgalpha = _mm_mullo_epi16(destalpha, alpha);
bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8); bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8);
__m128 fgalpha = _mm_mullo_epi16(srcalpha, alpha); __m128i fgalpha = _mm_mullo_epi16(srcalpha, alpha);
fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8); fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8);
fgcolor = _mm_mullo_epi16(fgcolor, fgalpha); fgcolor = _mm_mullo_epi16(fgcolor, fgalpha);
@ -783,15 +783,15 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8);
// Blend // Blend
__m128 alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); __m128i alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256 alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256
__m128 inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha); __m128i inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha);
__m128 bgalpha = _mm_mullo_epi16(destalpha, alpha); __m128i bgalpha = _mm_mullo_epi16(destalpha, alpha);
bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8); bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8);
__m128 fgalpha = _mm_mullo_epi16(srcalpha, alpha); __m128i fgalpha = _mm_mullo_epi16(srcalpha, alpha);
fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8); fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8);
fgcolor = _mm_mullo_epi16(fgcolor, fgalpha); fgcolor = _mm_mullo_epi16(fgcolor, fgalpha);
@ -873,15 +873,15 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8);
// Blend // Blend
__m128 alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); __m128i alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256 alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256
__m128 inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha); __m128i inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha);
__m128 bgalpha = _mm_mullo_epi16(destalpha, alpha); __m128i bgalpha = _mm_mullo_epi16(destalpha, alpha);
bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8); bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8);
__m128 fgalpha = _mm_mullo_epi16(srcalpha, alpha); __m128i fgalpha = _mm_mullo_epi16(srcalpha, alpha);
fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8); fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8);
fgcolor = _mm_mullo_epi16(fgcolor, fgalpha); fgcolor = _mm_mullo_epi16(fgcolor, fgalpha);
@ -976,15 +976,15 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8);
// Blend // Blend
__m128 alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); __m128i alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256 alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256
__m128 inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha); __m128i inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha);
__m128 bgalpha = _mm_mullo_epi16(destalpha, alpha); __m128i bgalpha = _mm_mullo_epi16(destalpha, alpha);
bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8); bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8);
__m128 fgalpha = _mm_mullo_epi16(srcalpha, alpha); __m128i fgalpha = _mm_mullo_epi16(srcalpha, alpha);
fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8); fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8);
fgcolor = _mm_mullo_epi16(fgcolor, fgalpha); fgcolor = _mm_mullo_epi16(fgcolor, fgalpha);
@ -1069,15 +1069,15 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8);
// Blend // Blend
__m128 alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); __m128i alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256 alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256
__m128 inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha); __m128i inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha);
__m128 bgalpha = _mm_mullo_epi16(destalpha, alpha); __m128i bgalpha = _mm_mullo_epi16(destalpha, alpha);
bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8); bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8);
__m128 fgalpha = _mm_mullo_epi16(srcalpha, alpha); __m128i fgalpha = _mm_mullo_epi16(srcalpha, alpha);
fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8); fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8);
fgcolor = _mm_mullo_epi16(fgcolor, fgalpha); fgcolor = _mm_mullo_epi16(fgcolor, fgalpha);
@ -1159,15 +1159,15 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8);
// Blend // Blend
__m128 alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); __m128i alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256 alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256
__m128 inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha); __m128i inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha);
__m128 bgalpha = _mm_mullo_epi16(destalpha, alpha); __m128i bgalpha = _mm_mullo_epi16(destalpha, alpha);
bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8); bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8);
__m128 fgalpha = _mm_mullo_epi16(srcalpha, alpha); __m128i fgalpha = _mm_mullo_epi16(srcalpha, alpha);
fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8); fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8);
fgcolor = _mm_mullo_epi16(fgcolor, fgalpha); fgcolor = _mm_mullo_epi16(fgcolor, fgalpha);
@ -1249,15 +1249,15 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8);
// Blend // Blend
__m128 alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); __m128i alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256 alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256
__m128 inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha); __m128i inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha);
__m128 bgalpha = _mm_mullo_epi16(destalpha, alpha); __m128i bgalpha = _mm_mullo_epi16(destalpha, alpha);
bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8); bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8);
__m128 fgalpha = _mm_mullo_epi16(srcalpha, alpha); __m128i fgalpha = _mm_mullo_epi16(srcalpha, alpha);
fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8); fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8);
fgcolor = _mm_mullo_epi16(fgcolor, fgalpha); fgcolor = _mm_mullo_epi16(fgcolor, fgalpha);
@ -1352,15 +1352,15 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8);
// Blend // Blend
__m128 alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); __m128i alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256 alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256
__m128 inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha); __m128i inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha);
__m128 bgalpha = _mm_mullo_epi16(destalpha, alpha); __m128i bgalpha = _mm_mullo_epi16(destalpha, alpha);
bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8); bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8);
__m128 fgalpha = _mm_mullo_epi16(srcalpha, alpha); __m128i fgalpha = _mm_mullo_epi16(srcalpha, alpha);
fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8); fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8);
fgcolor = _mm_mullo_epi16(fgcolor, fgalpha); fgcolor = _mm_mullo_epi16(fgcolor, fgalpha);
@ -1445,15 +1445,15 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8);
// Blend // Blend
__m128 alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); __m128i alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256 alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256
__m128 inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha); __m128i inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha);
__m128 bgalpha = _mm_mullo_epi16(destalpha, alpha); __m128i bgalpha = _mm_mullo_epi16(destalpha, alpha);
bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8); bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8);
__m128 fgalpha = _mm_mullo_epi16(srcalpha, alpha); __m128i fgalpha = _mm_mullo_epi16(srcalpha, alpha);
fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8); fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8);
fgcolor = _mm_mullo_epi16(fgcolor, fgalpha); fgcolor = _mm_mullo_epi16(fgcolor, fgalpha);
@ -1535,15 +1535,15 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, mlight), 8);
// Blend // Blend
__m128 alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); __m128i alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256 alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256
__m128 inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha); __m128i inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha);
__m128 bgalpha = _mm_mullo_epi16(destalpha, alpha); __m128i bgalpha = _mm_mullo_epi16(destalpha, alpha);
bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8); bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8);
__m128 fgalpha = _mm_mullo_epi16(srcalpha, alpha); __m128i fgalpha = _mm_mullo_epi16(srcalpha, alpha);
fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8); fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8);
fgcolor = _mm_mullo_epi16(fgcolor, fgalpha); fgcolor = _mm_mullo_epi16(fgcolor, fgalpha);
@ -1625,15 +1625,15 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8);
// Blend // Blend
__m128 alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); __m128i alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256 alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256
__m128 inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha); __m128i inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha);
__m128 bgalpha = _mm_mullo_epi16(destalpha, alpha); __m128i bgalpha = _mm_mullo_epi16(destalpha, alpha);
bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8); bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8);
__m128 fgalpha = _mm_mullo_epi16(srcalpha, alpha); __m128i fgalpha = _mm_mullo_epi16(srcalpha, alpha);
fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8); fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8);
fgcolor = _mm_mullo_epi16(fgcolor, fgalpha); fgcolor = _mm_mullo_epi16(fgcolor, fgalpha);
@ -1728,15 +1728,15 @@ namespace swrenderer
fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8); fgcolor = _mm_srli_epi16(_mm_mullo_epi16(fgcolor, shade_light), 8);
// Blend // Blend
__m128 alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); __m128i alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256 alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256
__m128 inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha); __m128i inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha);
__m128 bgalpha = _mm_mullo_epi16(destalpha, alpha); __m128i bgalpha = _mm_mullo_epi16(destalpha, alpha);
bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8); bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8);
__m128 fgalpha = _mm_mullo_epi16(srcalpha, alpha); __m128i fgalpha = _mm_mullo_epi16(srcalpha, alpha);
fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8); fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8);
fgcolor = _mm_mullo_epi16(fgcolor, fgalpha); fgcolor = _mm_mullo_epi16(fgcolor, fgalpha);

10
src/swrenderer/drawers/r_draw_wall32.php
View file

@ -198,7 +198,7 @@ namespace swrenderer
{ {
if ($blendVariant == "opaque") if ($blendVariant == "opaque")
{ ?> { ?>
__m128 outcolor = fgcolor; __m128i outcolor = fgcolor;
outcolor = _mm_packus_epi16(outcolor, _mm_setzero_si128()); outcolor = _mm_packus_epi16(outcolor, _mm_setzero_si128());
<? } <? }
else if ($blendVariant == "masked") else if ($blendVariant == "masked")
@ -244,20 +244,20 @@ namespace swrenderer
function CalcAlpha() function CalcAlpha()
{ ?> { ?>
__m128 alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); __m128i alpha = _mm_shufflelo_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3)); alpha = _mm_shufflehi_epi16(fgcolor, _MM_SHUFFLE(3,3,3,3));
alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256 alpha = _mm_add_epi16(alpha, _mm_srli_epi16(alpha, 7)); // 255 -> 256
__m128 inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha); __m128i inv_alpha = _mm_sub_epi16(_mm_set1_epi16(256), alpha);
<? } <? }
function CalcBlendColor() function CalcBlendColor()
{ {
CalcAlpha();?> CalcAlpha();?>
__m128 bgalpha = _mm_mullo_epi16(destalpha, alpha); __m128i bgalpha = _mm_mullo_epi16(destalpha, alpha);
bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8); bgalpha = _mm_srli_epi16(_mm_add_epi16(_mm_add_epi16(bgalpha, _mm_slli_epi16(inv_alpha, 8)), _mm_set1_epi16(128)), 8);
__m128 fgalpha = _mm_mullo_epi16(srcalpha, alpha); __m128i fgalpha = _mm_mullo_epi16(srcalpha, alpha);
fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8); fgalpha = _mm_srli_epi16(_mm_add_epi16(fgalpha, _mm_set1_epi16(128)), 8);
fgcolor = _mm_mullo_epi16(fgcolor, fgalpha); fgcolor = _mm_mullo_epi16(fgcolor, fgalpha);