More NEON MLA

hrydgard · Dec 9, 2023 · 08ce69d · 08ce69d
1 parent 55d4875
commit 08ce69d
Showing 1 changed file with 6 additions and 14 deletions.
diff --git a/GPU/Common/DrawEngineCommon.cpp b/GPU/Common/DrawEngineCommon.cpp
@@ -525,26 +525,18 @@ bool DrawEngineCommon::TestBoundingBoxFast(const void *vdata, int vertexCount, u
 		const float *pos = verts + i * vertStride;
 		float32x4_t objpos = vld1q_f32(pos);
 		float32x4_t worldpos = vaddq_f32(
-			vaddq_f32(
+			vmlaq_laneq_f32(
 				vmulq_laneq_f32(worldX, objpos, 0),
-				vmulq_laneq_f32(worldY, objpos, 1)
-			),
-			vaddq_f32(
-				vmulq_laneq_f32(worldZ, objpos, 2),
-				worldW
-			)
+				worldY, objpos, 1),
+			vmlaq_laneq_f32(worldW, worldZ, objpos, 2)
 		);
 		// OK, now we check it against the four planes.
 		// This is really curiously similar to a matrix multiplication (well, it is one).
 		float32x4_t planeDist = vaddq_f32(
-			vaddq_f32(
+			vmlaq_laneq_f32(
 				vmulq_laneq_f32(planeX, worldpos, 0),
-				vmulq_laneq_f32(planeY, worldpos, 1)
-			),
-			vaddq_f32(
-				vmulq_laneq_f32(planeZ, worldpos, 2),
-				planeW
-			)
+				planeY, worldpos, 1),
+			vmlaq_laneq_f32(planeW, planeZ, worldpos, 2)
 		);
 		inside = vorrq_u32(inside, vcgeq_f32(planeDist, vdupq_n_f32(0.0f)));
 	}