asrael/pxl8
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

cleanup some more f32 simd

Browse source
This commit is contained in:
asrael 2025-11-11 09:27:29 -06:00
parent 670cd3c97e
commit e2c7998663
2 changed files with 34 additions and 22 deletions
Download patch file Download diff file Expand all files Collapse all files

26
src/pxl8_gfx.c
View file

@ -810,29 +810,11 @@ void pxl8_3d_clear_zbuffer(pxl8_gfx* gfx) {
i32 count = gfx->zbuffer_width * gfx->zbuffer_height;
const f32 far_z = 1e30f;
#if defined(PXL8_SIMD_AVX2)
__m256 far_vec = _mm256_set1_ps(far_z);
#if !defined(PXL8_SIMD_SCALAR)
pxl8_simd_vec_f32 far_vec = pxl8_simd_set1_f32(far_z);
i32 i = 0;
for (; i + 7 < count; i += 8) {
_mm256_storeu_ps(&gfx->zbuffer[i], far_vec);
}
for (; i < count; i++) {
gfx->zbuffer[i] = far_z;
}
#elif defined(PXL8_SIMD_SSE2)
__m128 far_vec = _mm_set1_ps(far_z);
i32 i = 0;
for (; i + 3 < count; i += 4) {
_mm_storeu_ps(&gfx->zbuffer[i], far_vec);
}
for (; i < count; i++) {
gfx->zbuffer[i] = far_z;
}
#elif defined(PXL8_SIMD_NEON)
float32x4_t far_vec = vdupq_n_f32(far_z);
i32 i = 0;
for (; i + 3 < count; i += 4) {
vst1q_f32(&gfx->zbuffer[i], far_vec);
for (; i + PXL8_SIMD_WIDTH_F32 <= count; i += PXL8_SIMD_WIDTH_F32) {
pxl8_simd_store_f32(&gfx->zbuffer[i], far_vec);
}
for (; i < count; i++) {
gfx->zbuffer[i] = far_z;