krzosa/software_rasterizer
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

SIMD write to texture!

Browse Source
This commit is contained in:
Krzosa Karol
2022-07-06 14:15:59 +02:00
parent d2baefcc04
commit 1c3ff7e8cd
2 changed files with 23 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
build.bat
View File

@@ -6,4 +6,4 @@ rem assets.exe
rem tracy/TracyClient.cpp -DTRACY_ENABLE rem tracy/TracyClient.cpp -DTRACY_ENABLE
clang main.cpp -O2 -mavx2 -Wall -Wno-unused-function -Wno-missing-braces -fno-exceptions -fdiagnostics-absolute-paths -g -I".." -o main.exe -Wl,user32.lib -Wl,optick\lib\x64\release\OptickCore.lib clang main.cpp -mavx2 -Wall -Wno-unused-function -Wno-missing-braces -fno-exceptions -fdiagnostics-absolute-paths -g -I".." -o main.exe -Wl,user32.lib -Wl,optick\lib\x64\release\OptickCore.lib

34
main.cpp
View File

@@ -371,6 +371,8 @@ void draw_triangle_nearest(Bitmap* dst, F32 *depth_buffer, Bitmap *src, Vec3 lig
F32x8 one_over_p1w = _mm256_set1_ps(1.f / p1.w); F32x8 one_over_p1w = _mm256_set1_ps(1.f / p1.w);
F32x8 one_over_p2w = _mm256_set1_ps(1.f / p2.w); F32x8 one_over_p2w = _mm256_set1_ps(1.f / p2.w);
U32 *destination = dst->pixels + dst->x*min_y; U32 *destination = dst->pixels + dst->x*min_y;
F32 area = (p1.y - p0.y) * (p2.x - p0.x) - (p1.x - p0.x) * (p2.y - p0.y); F32 area = (p1.y - p0.y) * (p2.x - p0.x) - (p1.x - p0.x) * (p2.y - p0.y);
F32x8 area8 = _mm256_set1_ps(area); F32x8 area8 = _mm256_set1_ps(area);
@@ -423,7 +425,6 @@ void draw_triangle_nearest(Bitmap* dst, F32 *depth_buffer, Bitmap *src, Vec3 lig
F32 *depth_pointer = (depth_buffer + (x8 + y * dst->x)); F32 *depth_pointer = (depth_buffer + (x8 + y * dst->x));
F32x8 depth = _mm256_loadu_ps((float *)depth_pointer); F32x8 depth = _mm256_loadu_ps((float *)depth_pointer);
//
F32x8 should_fill_term = _mm256_cmp_ps(depth, interpolated_w, _CMP_LT_OQ); F32x8 should_fill_term = _mm256_cmp_ps(depth, interpolated_w, _CMP_LT_OQ);
should_fill = _mm256_and_ps(should_fill, should_fill_term); should_fill = _mm256_and_ps(should_fill, should_fill_term);
@@ -550,18 +551,27 @@ void draw_triangle_nearest(Bitmap* dst, F32 *depth_buffer, Bitmap *src, Vec3 lig
dst_b = _mm256_sqrt_ps(dst_b); dst_b = _mm256_sqrt_ps(dst_b);
} }
S32x8 result; // Convert to integer format
for(S64 i = 0; i < 8; i++){ dst_r = _mm256_mul_ps(dst_r, var255);
if (I(should_fill, i)){ dst_g = _mm256_mul_ps(dst_g, var255);
U8 red = (U8)(dst_r[i] * 255); dst_b = _mm256_mul_ps(dst_b, var255);
U8 green = (U8)(dst_g[i] * 255); dst_a = _mm256_mul_ps(dst_a, var255);
U8 blue = (U8)(dst_b[i] * 255);
U8 alpha = (U8)(dst_a[i] * 255);
Is(result, i) = (U32)(alpha << 24 | blue << 16 | green << 8 | red << 0);
}
}
_mm256_maskstore_epi32((int *)dst_memory, should_fill, result); S32x8 dst_r_int = _mm256_cvtps_epi32(dst_r);
S32x8 dst_g_int = _mm256_cvtps_epi32(dst_g);
S32x8 dst_b_int = _mm256_cvtps_epi32(dst_b);
S32x8 dst_a_int = _mm256_cvtps_epi32(dst_a);
S32x8 dst_int_a_shifted = _mm256_slli_epi32(dst_a_int, 24);
S32x8 dst_int_b_shifted = _mm256_slli_epi32(dst_b_int, 16);
S32x8 dst_int_g_shifted = _mm256_slli_epi32(dst_g_int, 8);
S32x8 dst_int_r_shifted = dst_r_int;
S32x8 packed_abgr0 = _mm256_or_si256(dst_int_a_shifted, dst_int_b_shifted);
S32x8 packed_abgr1 = _mm256_or_si256(packed_abgr0, dst_int_g_shifted);
S32x8 packed_abgr2 = _mm256_or_si256(packed_abgr1, dst_int_r_shifted);
_mm256_maskstore_epi32((int *)dst_memory, should_fill, packed_abgr2);
} }
Cy0 -= dx10; Cy0 -= dx10;