Allow i16 /⁼ to jump to Singeli code after range checking
This commit is contained in:
parent
926e59483d
commit
4aea668a6c
@ -826,7 +826,7 @@ B slash_im(B t, B x) {
|
|||||||
rp[sum>0] = sum; rp[0] = xia - sum;
|
rp[sum>0] = sum; rp[0] = xia - sum;
|
||||||
r = num_squeeze(r); break;
|
r = num_squeeze(r); break;
|
||||||
}
|
}
|
||||||
#define IIND_INT(N) \
|
#define IIND_INT(N, CHECK_RIA) \
|
||||||
if (xp[0]<0) thrM("/⁼: Argument cannot contain negative numbers"); \
|
if (xp[0]<0) thrM("/⁼: Argument cannot contain negative numbers"); \
|
||||||
usz a=1; while (a<xia && xp[a]>xp[a-1]) a++; \
|
usz a=1; while (a<xia && xp[a]>xp[a-1]) a++; \
|
||||||
u##N max=xp[a-1]; \
|
u##N max=xp[a-1]; \
|
||||||
@ -856,32 +856,35 @@ B slash_im(B t, B x) {
|
|||||||
break; \
|
break; \
|
||||||
} \
|
} \
|
||||||
usz ria = (usz)max + 1; \
|
usz ria = (usz)max + 1; \
|
||||||
|
CHECK_RIA \
|
||||||
i##N* rp; r = m_i##N##arrv(&rp, ria); for (usz i=0; i<ria; i++) rp[i]=0; \
|
i##N* rp; r = m_i##N##arrv(&rp, ria); for (usz i=0; i<ria; i++) rp[i]=0; \
|
||||||
for (usz i = 0; i < xia; i++) rp[xp[i]]++; \
|
for (usz i = 0; i < xia; i++) rp[xp[i]]++;
|
||||||
r = num_squeeze(r);
|
|
||||||
#define CASE_SMALL(N) \
|
#define CASE_SMALL(N) \
|
||||||
case el_i##N: { \
|
case el_i##N: { \
|
||||||
i##N* xp = i##N##any_ptr(x); \
|
i##N* xp = i##N##any_ptr(x); \
|
||||||
usz m=1<<N; \
|
usz m=1<<N; \
|
||||||
if (xia < m/2) { \
|
usz mh = m/2, sa = SINGELI_COUNT_ALLOC; \
|
||||||
IIND_INT(N) \
|
if (xia < mh) { \
|
||||||
|
IIND_INT(N, if (RIA_SMALL(N)) { sa=mh=ria; goto small_range##N; }) \
|
||||||
} else { \
|
} else { \
|
||||||
TALLOC(usz, t, SINGELI_COUNT_ALLOC); \
|
small_range##N: TALLOC(usz, t, sa); \
|
||||||
for (usz j=0; j<m/2; j++) t[j]=0; \
|
for (usz j=0; j<mh; j++) t[j]=0; \
|
||||||
SINGELI_COUNT(N) \
|
SINGELI_COUNT(N) \
|
||||||
i32* rp; r = m_i32arrv(&rp, ria); vfor (usz i=0; i<ria; i++) rp[i]=t[i]; \
|
i32* rp; r = m_i32arrv(&rp, ria); vfor (usz i=0; i<ria; i++) rp[i]=t[i]; \
|
||||||
TFREE(t); \
|
TFREE(t); \
|
||||||
r = num_squeeze(r); \
|
|
||||||
} \
|
} \
|
||||||
|
r = num_squeeze(r); \
|
||||||
break; \
|
break; \
|
||||||
}
|
}
|
||||||
#if SINGELI_SIMD
|
#if SINGELI_SIMD
|
||||||
|
#define RIA_SMALL(N) N==16 && ria<mh && ria+ria/2+64<=xia
|
||||||
#define SINGELI_COUNT_ALLOC m/2
|
#define SINGELI_COUNT_ALLOC m/2
|
||||||
#define SINGELI_COUNT(N) \
|
#define SINGELI_COUNT(N) \
|
||||||
i##N max = simd_count_i##N(t, xp, xia, 0); \
|
i##N max = simd_count_i##N(t, xp, xia, 0); \
|
||||||
if (max < 0) thrM("/⁼: Argument cannot contain negative numbers"); \
|
if (max < 0) thrM("/⁼: Argument cannot contain negative numbers"); \
|
||||||
usz ria=max+1;
|
usz ria=max+1;
|
||||||
#else
|
#else
|
||||||
|
#define RIA_SMALL(N) 0
|
||||||
#define SINGELI_COUNT_ALLOC m
|
#define SINGELI_COUNT_ALLOC m
|
||||||
#define SINGELI_COUNT(N) \
|
#define SINGELI_COUNT(N) \
|
||||||
for (usz i=0; i<xia; i++) t[(u##N)xp[i]]++; \
|
for (usz i=0; i<xia; i++) t[(u##N)xp[i]]++; \
|
||||||
@ -890,9 +893,10 @@ B slash_im(B t, B x) {
|
|||||||
#endif
|
#endif
|
||||||
CASE_SMALL(8) CASE_SMALL(16)
|
CASE_SMALL(8) CASE_SMALL(16)
|
||||||
#undef CASE_SMALL
|
#undef CASE_SMALL
|
||||||
|
#undef RIA_SMALL
|
||||||
#undef SINGELI_COUNT_ALLOC
|
#undef SINGELI_COUNT_ALLOC
|
||||||
#undef SINGELI_COUNT
|
#undef SINGELI_COUNT
|
||||||
case el_i32: { i32* xp = i32any_ptr(x); IIND_INT(32) r = num_squeeze(r); break; }
|
case el_i32: { i32* xp = i32any_ptr(x); IIND_INT(32,) r = num_squeeze(r); break; }
|
||||||
#undef IIND_INT
|
#undef IIND_INT
|
||||||
case el_f64: {
|
case el_f64: {
|
||||||
f64* xp = f64any_ptr(x);
|
f64* xp = f64any_ptr(x);
|
||||||
|
|||||||
Loading…
Reference in New Issue
Block a user