Skip to content

Commit 72a14e7

Browse files
9il9il
committed
partially fix spline
1 parent f3df9ef commit 72a14e7

File tree

2 files changed

+285
-182
lines changed

2 files changed

+285
-182
lines changed

‎source/mir/interpolate/package.d‎

Lines changed: 71 additions & 61 deletions
Original file line numberDiff line numberDiff line change
@@ -458,6 +458,16 @@ unittest
458458
assert(generateShuffles2!(4, 4) == [[0, 1, 2, 3], [4, 5, 6, 7]]);
459459
}
460460

461+
unittest
462+
{
463+
enum ai = [0, 1, 2, 3];
464+
enum bi = [4, 5, 6, 7];
465+
align(32)
466+
double[4] a = [0, 1, 2, 3], b = [4, 5, 6, 7], c, d;
467+
shuffle3!1(a, b, c, d);
468+
assert([c, d] == [[0.0, 4, 1, 5], [2.0, 6, 3, 7]]);
469+
}
470+
461471
unittest
462472
{
463473
enum ai = [0, 1, 2, 3];
@@ -490,78 +500,78 @@ import mir.internal.utility;
490500

491501
auto vectorize(Kernel, F, size_t N, size_t R)(ref Kernel kernel, ref F[N] a0, ref F[N] b0, ref F[N] a1, ref F[N] b1, ref F[N][R] c)
492502
{
493-
static if (LDC && F.mant_dig != 64)
494-
{
495-
alias V = __vector(F[N]); // @FUTURE@ vector support
496-
*cast(V[R]*) c.ptr = kernel(
497-
*cast(V*)a0.ptr, *cast(V*)b0.ptr,
498-
*cast(V*)a1.ptr, *cast(V*)b1.ptr);
499-
}
500-
else
501-
static if (F.sizeof <= double.sizeof && F[N].sizeof >= (double[2]).sizeof)
502-
{
503-
import mir.utility;
504-
enum S = _avx ? 32u : 16u;
505-
enum M = min(S, F[N].sizeof) / F.sizeof;
506-
alias V = __vector(F[M]); // @FUTURE@ vector support
507-
enum C = N / M;
508-
foreach(i; Iota!C)
509-
{
510-
auto r = kernel(
511-
*cast(V*)(a0.ptr + i * M), *cast(V*)(b0.ptr + i * M),
512-
*cast(V*)(a1.ptr + i * M), *cast(V*)(b1.ptr + i * M),
513-
);
514-
static if (R == 1)
515-
*cast(V*)(c[0].ptr + i * M) = r;
516-
else
517-
foreach(j; Iota!R)
518-
*cast(V*)(c[j].ptr + i * M) = r[j];
519-
}
520-
}
521-
else
522-
{
503+
// static if (LDC && F.mant_dig != 64)
504+
// {
505+
// alias V = __vector(F[N]); // @FUTURE@ vector support
506+
// *cast(V[R]*) c.ptr = kernel(
507+
// *cast(V*)a0.ptr, *cast(V*)b0.ptr,
508+
// *cast(V*)a1.ptr, *cast(V*)b1.ptr);
509+
// }
510+
// else
511+
// static if (F.sizeof <= double.sizeof && F[N].sizeof >= (double[2]).sizeof)
512+
// {
513+
// import mir.utility;
514+
// enum S = _avx ? 32u : 16u;
515+
// enum M = min(S, F[N].sizeof) / F.sizeof;
516+
// alias V = __vector(F[M]); // @FUTURE@ vector support
517+
// enum C = N / M;
518+
// foreach(i; Iota!C)
519+
// {
520+
// auto r = kernel(
521+
// *cast(V*)(a0.ptr + i * M), *cast(V*)(b0.ptr + i * M),
522+
// *cast(V*)(a1.ptr + i * M), *cast(V*)(b1.ptr + i * M),
523+
// );
524+
// static if (R == 1)
525+
// *cast(V*)(c[0].ptr + i * M) = r;
526+
// else
527+
// foreach(j; Iota!R)
528+
// *cast(V*)(c[j].ptr + i * M) = r[j];
529+
// }
530+
// }
531+
// else
532+
// {
523533
foreach(i; Iota!N)
524534
{
525535
auto r = kernel(a0[i], b0[i], a1[i], b1[i]);
526536
static if (R == 1)
527-
return c[0] = r;
537+
c[0][i] = r;
528538
else
529539
foreach(j; Iota!R)
530540
c[j][i] = r[j];
531541
}
532-
}
542+
// }
533543
}
534544

535545
auto vectorize(Kernel, F, size_t N, size_t R)(ref Kernel kernel, ref F[N] a, ref F[N] b, ref F[N][R] c)
536546
{
537-
static if (LDC && F.mant_dig != 64 && is(__vector(F[N])))
538-
{
539-
alias V = __vector(F[N]); // @FUTURE@ vector support
540-
*cast(V[R]*) c.ptr = kernel(*cast(V*)a.ptr, *cast(V*)b.ptr);
541-
}
542-
else
543-
static if (F.sizeof <= double.sizeof && F[N].sizeof >= (double[2]).sizeof && x86_64)
544-
{
545-
import mir.utility;
546-
enum S = _avx ? 32u : 16u;
547-
enum M = min(S, F[N].sizeof) / F.sizeof;
548-
alias V = __vector(F[M]); // @FUTURE@ vector support
549-
enum C = N / M;
550-
foreach(i; Iota!C)
551-
{
552-
auto r = kernel(
553-
*cast(V*)(a.ptr + i * M),
554-
*cast(V*)(b.ptr + i * M),
555-
);
556-
static if (R == 1)
557-
*cast(V*)(c[0].ptr + i * M) = r;
558-
else
559-
foreach(j; Iota!R)
560-
*cast(V*)(c[j].ptr + i * M) = r[j];
561-
}
562-
}
563-
else
564-
{
547+
// static if (LDC && F.mant_dig != 64 && is(__vector(F[N])))
548+
// {
549+
// alias V = __vector(F[N]); // @FUTURE@ vector support
550+
// *cast(V[R]*) c.ptr = kernel(*cast(V*)a.ptr, *cast(V*)b.ptr);
551+
// }
552+
// else
553+
// static if (F.sizeof <= double.sizeof && F[N].sizeof >= (double[2]).sizeof && x86_64)
554+
// {
555+
// import mir.utility;
556+
// enum S = _avx ? 32u : 16u;
557+
// enum M = min(S, F[N].sizeof) / F.sizeof;
558+
// alias V = __vector(F[M]); // @FUTURE@ vector support
559+
// enum C = N / M;
560+
// foreach(i; Iota!C)
561+
// {
562+
// auto r = kernel(
563+
// *cast(V*)(a.ptr + i * M),
564+
// *cast(V*)(b.ptr + i * M),
565+
// );
566+
// static if (R == 1)
567+
// *cast(V*)(c[0].ptr + i * M) = r;
568+
// else
569+
// foreach(j; Iota!R)
570+
// *cast(V*)(c[j].ptr + i * M) = r[j];
571+
// }
572+
// }
573+
// else
574+
// {
565575
F[N][R] _c = void;//Temporary array in case "c" overlaps "a" and/or "b".
566576
foreach(i; Iota!N)
567577
{
@@ -573,7 +583,7 @@ auto vectorize(Kernel, F, size_t N, size_t R)(ref Kernel kernel, ref F[N] a, ref
573583
_c[j][i] = r[j];
574584
}
575585
c = _c;
576-
}
586+
// }
577587
}
578588

579589
// version(unittest)

0 commit comments

Comments
 (0)