Add and use SSE-enabled ImRsqrt() in place of 1.0f / ImSqrt(). (#4091)

Squashed 3 commits.
This commit is contained in:
Bartosz Taudul
2021-05-06 17:25:40 +02:00
committed by ocornut
parent 84545dbe6f
commit 4c9f0cec27
3 changed files with 15 additions and 2 deletions

View File

@ -694,7 +694,7 @@ void ImDrawList::PrimQuadUV(const ImVec2& a, const ImVec2& b, const ImVec2& c, c
// On AddPolyline() and AddConvexPolyFilled() we intentionally avoid using ImVec2 and superfluous function calls to optimize debug/non-inlined builds.
// Those macros expects l-values.
#define IM_NORMALIZE2F_OVER_ZERO(VX,VY) do { float d2 = VX*VX + VY*VY; if (d2 > 0.0f) { float inv_len = 1.0f / ImSqrt(d2); VX *= inv_len; VY *= inv_len; } } while (0)
#define IM_NORMALIZE2F_OVER_ZERO(VX,VY) do { float d2 = VX*VX + VY*VY; if (d2 > 0.0f) { float inv_len = ImRsqrt(d2); VX *= inv_len; VY *= inv_len; } } while (0)
#define IM_FIXNORMAL2F_MAX_INVLEN2 100.0f // 500.0f (see #4053, #3366)
#define IM_FIXNORMAL2F(VX,VY) do { float d2 = VX*VX + VY*VY; if (d2 > 0.000001f) { float inv_len2 = 1.0f / d2; if (inv_len2 > IM_FIXNORMAL2F_MAX_INVLEN2) inv_len2 = IM_FIXNORMAL2F_MAX_INVLEN2; VX *= inv_len2; VY *= inv_len2; } } while (0)