
(FPCore (normAngle u n0_i n1_i)
:precision binary32
(let* ((t_0 (/ 1.0 (sin normAngle))))
(+
(* (* (sin (* (- 1.0 u) normAngle)) t_0) n0_i)
(* (* (sin (* u normAngle)) t_0) n1_i))))
float code(float normAngle, float u, float n0_i, float n1_i) {
float t_0 = 1.0f / sinf(normAngle);
return ((sinf(((1.0f - u) * normAngle)) * t_0) * n0_i) + ((sinf((u * normAngle)) * t_0) * n1_i);
}
module fmin_fmax_functions
implicit none
private
public fmax
public fmin
interface fmax
module procedure fmax88
module procedure fmax44
module procedure fmax84
module procedure fmax48
end interface
interface fmin
module procedure fmin88
module procedure fmin44
module procedure fmin84
module procedure fmin48
end interface
contains
real(8) function fmax88(x, y) result (res)
real(8), intent (in) :: x
real(8), intent (in) :: y
res = merge(y, merge(x, max(x, y), y /= y), x /= x)
end function
real(4) function fmax44(x, y) result (res)
real(4), intent (in) :: x
real(4), intent (in) :: y
res = merge(y, merge(x, max(x, y), y /= y), x /= x)
end function
real(8) function fmax84(x, y) result(res)
real(8), intent (in) :: x
real(4), intent (in) :: y
res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
end function
real(8) function fmax48(x, y) result(res)
real(4), intent (in) :: x
real(8), intent (in) :: y
res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
end function
real(8) function fmin88(x, y) result (res)
real(8), intent (in) :: x
real(8), intent (in) :: y
res = merge(y, merge(x, min(x, y), y /= y), x /= x)
end function
real(4) function fmin44(x, y) result (res)
real(4), intent (in) :: x
real(4), intent (in) :: y
res = merge(y, merge(x, min(x, y), y /= y), x /= x)
end function
real(8) function fmin84(x, y) result(res)
real(8), intent (in) :: x
real(4), intent (in) :: y
res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
end function
real(8) function fmin48(x, y) result(res)
real(4), intent (in) :: x
real(8), intent (in) :: y
res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
end function
end module
real(4) function code(normangle, u, n0_i, n1_i)
use fmin_fmax_functions
real(4), intent (in) :: normangle
real(4), intent (in) :: u
real(4), intent (in) :: n0_i
real(4), intent (in) :: n1_i
real(4) :: t_0
t_0 = 1.0e0 / sin(normangle)
code = ((sin(((1.0e0 - u) * normangle)) * t_0) * n0_i) + ((sin((u * normangle)) * t_0) * n1_i)
end function
function code(normAngle, u, n0_i, n1_i) t_0 = Float32(Float32(1.0) / sin(normAngle)) return Float32(Float32(Float32(sin(Float32(Float32(Float32(1.0) - u) * normAngle)) * t_0) * n0_i) + Float32(Float32(sin(Float32(u * normAngle)) * t_0) * n1_i)) end
function tmp = code(normAngle, u, n0_i, n1_i) t_0 = single(1.0) / sin(normAngle); tmp = ((sin(((single(1.0) - u) * normAngle)) * t_0) * n0_i) + ((sin((u * normAngle)) * t_0) * n1_i); end
\begin{array}{l}
\\
\begin{array}{l}
t_0 := \frac{1}{\sin normAngle}\\
\left(\sin \left(\left(1 - u\right) \cdot normAngle\right) \cdot t\_0\right) \cdot n0\_i + \left(\sin \left(u \cdot normAngle\right) \cdot t\_0\right) \cdot n1\_i
\end{array}
\end{array}
Sampling outcomes in binary32 precision:
Herbie found 2 alternatives:
| Alternative | Accuracy | Speedup |
|---|
(FPCore (normAngle u n0_i n1_i)
:precision binary32
(let* ((t_0 (/ 1.0 (sin normAngle))))
(+
(* (* (sin (* (- 1.0 u) normAngle)) t_0) n0_i)
(* (* (sin (* u normAngle)) t_0) n1_i))))
float code(float normAngle, float u, float n0_i, float n1_i) {
float t_0 = 1.0f / sinf(normAngle);
return ((sinf(((1.0f - u) * normAngle)) * t_0) * n0_i) + ((sinf((u * normAngle)) * t_0) * n1_i);
}
module fmin_fmax_functions
implicit none
private
public fmax
public fmin
interface fmax
module procedure fmax88
module procedure fmax44
module procedure fmax84
module procedure fmax48
end interface
interface fmin
module procedure fmin88
module procedure fmin44
module procedure fmin84
module procedure fmin48
end interface
contains
real(8) function fmax88(x, y) result (res)
real(8), intent (in) :: x
real(8), intent (in) :: y
res = merge(y, merge(x, max(x, y), y /= y), x /= x)
end function
real(4) function fmax44(x, y) result (res)
real(4), intent (in) :: x
real(4), intent (in) :: y
res = merge(y, merge(x, max(x, y), y /= y), x /= x)
end function
real(8) function fmax84(x, y) result(res)
real(8), intent (in) :: x
real(4), intent (in) :: y
res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
end function
real(8) function fmax48(x, y) result(res)
real(4), intent (in) :: x
real(8), intent (in) :: y
res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
end function
real(8) function fmin88(x, y) result (res)
real(8), intent (in) :: x
real(8), intent (in) :: y
res = merge(y, merge(x, min(x, y), y /= y), x /= x)
end function
real(4) function fmin44(x, y) result (res)
real(4), intent (in) :: x
real(4), intent (in) :: y
res = merge(y, merge(x, min(x, y), y /= y), x /= x)
end function
real(8) function fmin84(x, y) result(res)
real(8), intent (in) :: x
real(4), intent (in) :: y
res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
end function
real(8) function fmin48(x, y) result(res)
real(4), intent (in) :: x
real(8), intent (in) :: y
res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
end function
end module
real(4) function code(normangle, u, n0_i, n1_i)
use fmin_fmax_functions
real(4), intent (in) :: normangle
real(4), intent (in) :: u
real(4), intent (in) :: n0_i
real(4), intent (in) :: n1_i
real(4) :: t_0
t_0 = 1.0e0 / sin(normangle)
code = ((sin(((1.0e0 - u) * normangle)) * t_0) * n0_i) + ((sin((u * normangle)) * t_0) * n1_i)
end function
function code(normAngle, u, n0_i, n1_i) t_0 = Float32(Float32(1.0) / sin(normAngle)) return Float32(Float32(Float32(sin(Float32(Float32(Float32(1.0) - u) * normAngle)) * t_0) * n0_i) + Float32(Float32(sin(Float32(u * normAngle)) * t_0) * n1_i)) end
function tmp = code(normAngle, u, n0_i, n1_i) t_0 = single(1.0) / sin(normAngle); tmp = ((sin(((single(1.0) - u) * normAngle)) * t_0) * n0_i) + ((sin((u * normAngle)) * t_0) * n1_i); end
\begin{array}{l}
\\
\begin{array}{l}
t_0 := \frac{1}{\sin normAngle}\\
\left(\sin \left(\left(1 - u\right) \cdot normAngle\right) \cdot t\_0\right) \cdot n0\_i + \left(\sin \left(u \cdot normAngle\right) \cdot t\_0\right) \cdot n1\_i
\end{array}
\end{array}
(FPCore (normAngle u n0_i n1_i)
:precision binary32
(let* ((t_0 (fma -0.027777777777777776 n1_i (* 0.008333333333333333 n1_i)))
(t_1 (- (* -0.5 n0_i) (* -0.16666666666666666 n0_i)))
(t_2
(-
(* 0.041666666666666664 n0_i)
(fma -0.16666666666666666 t_1 (* 0.008333333333333333 n0_i)))))
(fma
(+
n1_i
(fma
-1.0
n0_i
(*
(* normAngle normAngle)
(-
(fma
-1.0
t_1
(fma
u
(fma
-0.5
n0_i
(* u (fma -0.16666666666666666 n1_i (* 0.16666666666666666 n0_i))))
(*
(* normAngle normAngle)
(-
(fma
-1.0
t_2
(fma
(* normAngle normAngle)
(-
(fma
-1.0
(-
(* -0.001388888888888889 n0_i)
(+
(* -0.16666666666666666 t_2)
(fma
-0.0001984126984126984
n0_i
(* 0.008333333333333333 t_1))))
(*
(* u u)
(fma 0.16666666666666666 t_0 (* 0.16666666666666666 t_2))))
(fma
-0.0001984126984126984
n1_i
(fma 0.001388888888888889 n1_i (* 0.16666666666666666 t_0))))
(*
(* u u)
(fma -0.027777777777777776 n1_i (* 0.16666666666666666 t_1)))))
t_0))))
(* -0.16666666666666666 n1_i)))))
u
n0_i)))
float code(float normAngle, float u, float n0_i, float n1_i) {
float t_0 = fmaf(-0.027777777777777776f, n1_i, (0.008333333333333333f * n1_i));
float t_1 = (-0.5f * n0_i) - (-0.16666666666666666f * n0_i);
float t_2 = (0.041666666666666664f * n0_i) - fmaf(-0.16666666666666666f, t_1, (0.008333333333333333f * n0_i));
return fmaf((n1_i + fmaf(-1.0f, n0_i, ((normAngle * normAngle) * (fmaf(-1.0f, t_1, fmaf(u, fmaf(-0.5f, n0_i, (u * fmaf(-0.16666666666666666f, n1_i, (0.16666666666666666f * n0_i)))), ((normAngle * normAngle) * (fmaf(-1.0f, t_2, fmaf((normAngle * normAngle), (fmaf(-1.0f, ((-0.001388888888888889f * n0_i) - ((-0.16666666666666666f * t_2) + fmaf(-0.0001984126984126984f, n0_i, (0.008333333333333333f * t_1)))), ((u * u) * fmaf(0.16666666666666666f, t_0, (0.16666666666666666f * t_2)))) - fmaf(-0.0001984126984126984f, n1_i, fmaf(0.001388888888888889f, n1_i, (0.16666666666666666f * t_0)))), ((u * u) * fmaf(-0.027777777777777776f, n1_i, (0.16666666666666666f * t_1))))) - t_0)))) - (-0.16666666666666666f * n1_i))))), u, n0_i);
}
function code(normAngle, u, n0_i, n1_i) t_0 = fma(Float32(-0.027777777777777776), n1_i, Float32(Float32(0.008333333333333333) * n1_i)) t_1 = Float32(Float32(Float32(-0.5) * n0_i) - Float32(Float32(-0.16666666666666666) * n0_i)) t_2 = Float32(Float32(Float32(0.041666666666666664) * n0_i) - fma(Float32(-0.16666666666666666), t_1, Float32(Float32(0.008333333333333333) * n0_i))) return fma(Float32(n1_i + fma(Float32(-1.0), n0_i, Float32(Float32(normAngle * normAngle) * Float32(fma(Float32(-1.0), t_1, fma(u, fma(Float32(-0.5), n0_i, Float32(u * fma(Float32(-0.16666666666666666), n1_i, Float32(Float32(0.16666666666666666) * n0_i)))), Float32(Float32(normAngle * normAngle) * Float32(fma(Float32(-1.0), t_2, fma(Float32(normAngle * normAngle), Float32(fma(Float32(-1.0), Float32(Float32(Float32(-0.001388888888888889) * n0_i) - Float32(Float32(Float32(-0.16666666666666666) * t_2) + fma(Float32(-0.0001984126984126984), n0_i, Float32(Float32(0.008333333333333333) * t_1)))), Float32(Float32(u * u) * fma(Float32(0.16666666666666666), t_0, Float32(Float32(0.16666666666666666) * t_2)))) - fma(Float32(-0.0001984126984126984), n1_i, fma(Float32(0.001388888888888889), n1_i, Float32(Float32(0.16666666666666666) * t_0)))), Float32(Float32(u * u) * fma(Float32(-0.027777777777777776), n1_i, Float32(Float32(0.16666666666666666) * t_1))))) - t_0)))) - Float32(Float32(-0.16666666666666666) * n1_i))))), u, n0_i) end
\begin{array}{l}
\\
\begin{array}{l}
t_0 := \mathsf{fma}\left(-0.027777777777777776, n1\_i, 0.008333333333333333 \cdot n1\_i\right)\\
t_1 := -0.5 \cdot n0\_i - -0.16666666666666666 \cdot n0\_i\\
t_2 := 0.041666666666666664 \cdot n0\_i - \mathsf{fma}\left(-0.16666666666666666, t\_1, 0.008333333333333333 \cdot n0\_i\right)\\
\mathsf{fma}\left(n1\_i + \mathsf{fma}\left(-1, n0\_i, \left(normAngle \cdot normAngle\right) \cdot \left(\mathsf{fma}\left(-1, t\_1, \mathsf{fma}\left(u, \mathsf{fma}\left(-0.5, n0\_i, u \cdot \mathsf{fma}\left(-0.16666666666666666, n1\_i, 0.16666666666666666 \cdot n0\_i\right)\right), \left(normAngle \cdot normAngle\right) \cdot \left(\mathsf{fma}\left(-1, t\_2, \mathsf{fma}\left(normAngle \cdot normAngle, \mathsf{fma}\left(-1, -0.001388888888888889 \cdot n0\_i - \left(-0.16666666666666666 \cdot t\_2 + \mathsf{fma}\left(-0.0001984126984126984, n0\_i, 0.008333333333333333 \cdot t\_1\right)\right), \left(u \cdot u\right) \cdot \mathsf{fma}\left(0.16666666666666666, t\_0, 0.16666666666666666 \cdot t\_2\right)\right) - \mathsf{fma}\left(-0.0001984126984126984, n1\_i, \mathsf{fma}\left(0.001388888888888889, n1\_i, 0.16666666666666666 \cdot t\_0\right)\right), \left(u \cdot u\right) \cdot \mathsf{fma}\left(-0.027777777777777776, n1\_i, 0.16666666666666666 \cdot t\_1\right)\right)\right) - t\_0\right)\right)\right) - -0.16666666666666666 \cdot n1\_i\right)\right), u, n0\_i\right)
\end{array}
\end{array}
Initial program 97.8%
Taylor expanded in u around 0
Applied rewrites95.2%
Taylor expanded in normAngle around 0
Applied rewrites99.5%
(FPCore (normAngle u n0_i n1_i) :precision binary32 (* (fma (/ n0_i n1_i) (/ (sin (* (- 1.0 u) normAngle)) (sin normAngle)) (/ (sin (* normAngle u)) (sin normAngle))) n1_i))
float code(float normAngle, float u, float n0_i, float n1_i) {
return fmaf((n0_i / n1_i), (sinf(((1.0f - u) * normAngle)) / sinf(normAngle)), (sinf((normAngle * u)) / sinf(normAngle))) * n1_i;
}
function code(normAngle, u, n0_i, n1_i) return Float32(fma(Float32(n0_i / n1_i), Float32(sin(Float32(Float32(Float32(1.0) - u) * normAngle)) / sin(normAngle)), Float32(sin(Float32(normAngle * u)) / sin(normAngle))) * n1_i) end
\begin{array}{l}
\\
\mathsf{fma}\left(\frac{n0\_i}{n1\_i}, \frac{\sin \left(\left(1 - u\right) \cdot normAngle\right)}{\sin normAngle}, \frac{\sin \left(normAngle \cdot u\right)}{\sin normAngle}\right) \cdot n1\_i
\end{array}
Initial program 97.8%
Taylor expanded in n1_i around inf
*-commutativeN/A
lower-*.f32N/A
Applied rewrites98.0%
herbie shell --seed 2025065
(FPCore (normAngle u n0_i n1_i)
:name "Curve intersection, scale width based on ribbon orientation"
:precision binary32
:pre (and (and (and (and (<= 0.0 normAngle) (<= normAngle (/ PI 2.0))) (and (<= -1.0 n0_i) (<= n0_i 1.0))) (and (<= -1.0 n1_i) (<= n1_i 1.0))) (and (<= 2.328306437e-10 u) (<= u 1.0)))
(+ (* (* (sin (* (- 1.0 u) normAngle)) (/ 1.0 (sin normAngle))) n0_i) (* (* (sin (* u normAngle)) (/ 1.0 (sin normAngle))) n1_i)))