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

Before adding FMA

Browse Source
This commit is contained in:
Krzosa Karol
2022-07-08 22:08:24 +02:00
parent f70cfd9bf5
commit 6e8d4b0137
2 changed files with 5 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
build.bat
View File

@@ -1,5 +1,5 @@
@echo off @echo off
pushd %~dp0 pushd %~dp0
clang main.cpp -mavx2 -Wall -Wno-unused-function -Wno-missing-braces -fno-exceptions -fdiagnostics-absolute-paths -I".." -g -o main.exe -Wl,user32.lib clang main.cpp -O2 -mfma -mavx2 -Wall -Wno-unused-function -Wno-missing-braces -fno-exceptions -fdiagnostics-absolute-paths -I".." -g -o main.exe -Wl,user32.lib
popd popd

8
main.cpp
View File

@@ -364,18 +364,17 @@ void draw_triangle_nearest(Bitmap* dst, F32 *depth_buffer, Bitmap *src, Vec3 lig
U64 fill_pixels_begin = __rdtsc(); U64 fill_pixels_begin = __rdtsc();
for (S64 y = min_y; y < max_y; y++) { for (S64 y = min_y; y < max_y; y++) {
F32x8 Y = _mm256_set1_ps(y); F32x8 Y = _mm256_set1_ps(y);
for (S64 x8 = min_x; x8 < max_x; x8+=8) { for (S64 x8 = min_x; x8 < max_x; x8+=8) {
F32x8 X = _mm256_add_ps(_mm256_set1_ps(x8), var07); F32x8 X = _mm256_add_ps(_mm256_set1_ps(x8), var07);
// Compute the edges // Compute the edges
// F32x8 edge0 = (p1.y - p0.y) * (p.x - p0.x) - (p1.x - p0.x) * (p.y - p0.y); // F32x8 edge0 = (p1.y - p0.y) * (p.x - p0.x) - (p1.x - p0.x) * (p.y - p0.y);
F32x8 px_minus_0x = _mm256_sub_ps(X, p0_x); F32x8 px_minus_0x = _mm256_sub_ps(X, p0_x);
F32x8 py_minus_0y = _mm256_sub_ps(Y, p0_y); F32x8 py_minus_0y = _mm256_sub_ps(Y, p0_y);
F32x8 left0 = _mm256_mul_ps(_dy10, px_minus_0x); F32x8 left0 = _mm256_mul_ps(_dy10, px_minus_0x);
F32x8 right0 = _mm256_mul_ps(_dx10, py_minus_0y); F32x8 right0 = _mm256_mul_ps(_dx10, py_minus_0y);
// F32x8 edge0 = _mm256_fmsub_ps(_dy10, px_minus_0x, right0);
F32x8 edge0 = _mm256_sub_ps(left0,right0); F32x8 edge0 = _mm256_sub_ps(left0,right0);
// F32 result = (p2.y - p1.y) * (p.x - p1.x) - (p2.x - p1.x) * (p.y - p1.y); // F32 result = (p2.y - p1.y) * (p.x - p1.x) - (p2.x - p1.x) * (p.y - p1.y);
@@ -1069,8 +1068,9 @@ test_array_list(){
array_add(scratch, &array, 31); array_add(scratch, &array, 31);
array_add(scratch, &array, 32); array_add(scratch, &array, 32);
array_ordered_remove(&array, 32); array_ordered_remove(&array, 32);
array_ordered_remove(&array, 16);
array_ordered_remove(&array, 0); array_ordered_remove(&array, 0);
array_ordered_remove(&array, 16);
array_ordered_remove(&array, 29);
array_print(&array); array_print(&array);
} }
@@ -1102,7 +1102,7 @@ test_array_list(){
array_print(&array); array_print(&array);
} }
__debugbreak(); // __debugbreak();
} }
FILE *global_file; FILE *global_file;