preserve-emulation/ppsspp
1
0 Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2026-10-11 05:26:24 +02:00
Code Issues Packages Projects Releases Wiki Activity

NEON: vcvtq can scale directly, no need for a mul by const.

Browse Source
This commit is contained in:
Henrik Rydgård committed 2023-12-09 16:48:59 +01:00
1 parent 99548be8a3
commit 4e2a1bf81c
2 files changed
+7 -4
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

GPU/Common/DrawEngineCommon.cpp
+1 -2
View File
@@ -444,11 +444,10 @@ bool DrawEngineCommon::TestBoundingBoxFast(const void *vdata, int vertexCount, u
_mm_storeu_ps(verts + i * 3, pos); // TODO: use stride 4 to avoid clashing writes?
}
#elif PPSSPP_ARCH(ARM_NEON)
float32x4_t scaleFactor = vdupq_n_f32(1.0f / 32768.0f);
for (int i = 0; i < vertexCount; i++) {
const s16 *dataPtr = ((const s16 *)((const s8 *)vdata + i * stride + offset));
int32x4_t data = vmovl_s16(vld1_s16(dataPtr));
float32x4_t pos = vmulq_f32(vcvtq_f32_s32(data), scaleFactor);
float32x4_t pos = vcvtq_n_f32_s32(data, 15); // >> 15 = division by 32768.0f
vst1q_f32(verts + i * 3, pos);
}
#else