diff options
author | Quincey Koziol <koziol@hdfgroup.org> | 2004-12-29 14:26:20 (GMT) |
---|---|---|
committer | Quincey Koziol <koziol@hdfgroup.org> | 2004-12-29 14:26:20 (GMT) |
commit | 427ff7da2848042f68ecfadf5a321b1d8077e9db (patch) | |
tree | 73024b1954031fbb724c2d96a485590348e5cc22 /src/H5V.c | |
parent | 9b96fd2003ae74cca389cc4c2216b4371d6eb173 (diff) | |
download | hdf5-427ff7da2848042f68ecfadf5a321b1d8077e9db.zip hdf5-427ff7da2848042f68ecfadf5a321b1d8077e9db.tar.gz hdf5-427ff7da2848042f68ecfadf5a321b1d8077e9db.tar.bz2 |
[svn-r9727] Purpose:
Bug Fix/Code Cleanup/Doc Cleanup/Optimization/Branch Sync :-)
Description:
Generally speaking, this is the "signed->unsigned" change to selections.
However, in the process of merging code back, things got stickier and stickier
until I ended up doing a big "sync the two branches up" operation. So... I
brought back all the "infrastructure" fixes from the development branch to the
release branch (which I think were actually making some improvement in
performance) as well as fixed several bugs which had been fixed in one branch,
but not the other.
I've also tagged the repository before making this checkin with the label
"before_signed_unsigned_changes".
Platforms tested:
FreeBSD 4.10 (sleipnir) w/parallel & fphdf5
FreeBSD 4.10 (sleipnir) w/threadsafe
FreeBSD 4.10 (sleipnir) w/backward compatibility
Solaris 2.7 (arabica) w/"purify options"
Solaris 2.8 (sol) w/FORTRAN & C++
AIX 5.x (copper) w/parallel & FORTRAN
IRIX64 6.5 (modi4) w/FORTRAN
Linux 2.4 (heping) w/FORTRAN & C++
Misc. update:
Diffstat (limited to 'src/H5V.c')
-rw-r--r-- | src/H5V.c | 390 |
1 files changed, 204 insertions, 186 deletions
@@ -32,15 +32,15 @@ /* Local prototypes */ static void H5V_stride_optimize1(unsigned *np/*in,out*/, hsize_t *elmt_size/*in,out*/, - const hsize_t *size, hssize_t *stride1); + const hsize_t *size, hsize_t *stride1); static void H5V_stride_optimize2(unsigned *np/*in,out*/, hsize_t *elmt_size/*in,out*/, - const hsize_t *size, hssize_t *stride1, hssize_t *stride2); + const hsize_t *size, hsize_t *stride1, hsize_t *stride2); #ifdef LATER static void H5V_stride_copy2(hsize_t nelmts, hsize_t elmt_size, - unsigned dst_n, const hsize_t *dst_size, const hssize_t *dst_stride, void *_dst, - unsigned src_n, const hsize_t *src_size, const hssize_t *src_stride, const void *_src); + unsigned dst_n, const hsize_t *dst_size, const ssize_t *dst_stride, void *_dst, + unsigned src_n, const hsize_t *src_size, const ssize_t *src_stride, const void *_src); #endif /* LATER */ @@ -66,7 +66,7 @@ H5V_stride_copy2(hsize_t nelmts, hsize_t elmt_size, */ static void H5V_stride_optimize1(unsigned *np/*in,out*/, hsize_t *elmt_size/*in,out*/, - const hsize_t *size, hssize_t *stride1) + const hsize_t *size, hsize_t *stride1) { FUNC_ENTER_NOAPI_NOINIT_NOFUNC(H5V_stride_optimize1); @@ -82,10 +82,8 @@ H5V_stride_optimize1(unsigned *np/*in,out*/, hsize_t *elmt_size/*in,out*/, while (*np && stride1[*np-1]>0 && (hsize_t)(stride1[*np-1])==*elmt_size) { *elmt_size *= size[*np-1]; - if (--*np) { - H5_CHECK_OVERFLOW(size[*np],hsize_t,hssize_t); - stride1[*np-1] += (hssize_t)size[*np] * stride1[*np]; - } + if (--*np) + stride1[*np-1] += size[*np] * stride1[*np]; } FUNC_LEAVE_NOAPI_VOID @@ -117,7 +115,7 @@ H5V_stride_optimize1(unsigned *np/*in,out*/, hsize_t *elmt_size/*in,out*/, */ static void H5V_stride_optimize2(unsigned *np/*in,out*/, hsize_t *elmt_size/*in,out*/, - const hsize_t *size, hssize_t *stride1, hssize_t *stride2) + const hsize_t *size, hsize_t *stride1, hsize_t *stride2) { FUNC_ENTER_NOAPI_NOINIT_NOFUNC(H5V_stride_optimize2) @@ -135,24 +133,20 @@ H5V_stride_optimize2(unsigned *np/*in,out*/, hsize_t *elmt_size/*in,out*/, /* Unroll loop for common cases */ switch(*np) { case 1: /* For 0-D datasets (dunno if this ever gets used...) */ - if((hsize_t)(stride1[0]) == *elmt_size && - (hsize_t)(stride2[0]) == *elmt_size) { + if(stride1[0] == *elmt_size && stride2[0] == *elmt_size) { *elmt_size *= size[0]; --*np; /* *np decrements to a value of 0 now */ } /* end if */ break; case 2: /* For 1-D datasets */ - if((hsize_t)(stride1[1]) == *elmt_size && - (hsize_t)(stride2[1]) == *elmt_size) { + if(stride1[1] == *elmt_size && stride2[1] == *elmt_size) { *elmt_size *= size[1]; --*np; /* *np decrements to a value of 1 now */ - H5_CHECK_OVERFLOW(size[1],hsize_t,hssize_t); - stride1[0] += (hssize_t)size[1] * stride1[1]; - stride2[0] += (hssize_t)size[1] * stride2[1]; + stride1[0] += size[1] * stride1[1]; + stride2[0] += size[1] * stride2[1]; - if((hsize_t)(stride1[0]) == *elmt_size && - (hsize_t)(stride2[0]) == *elmt_size) { + if(stride1[0] == *elmt_size && stride2[0] == *elmt_size) { *elmt_size *= size[0]; --*np; /* *np decrements to a value of 0 now */ } /* end if */ @@ -160,24 +154,19 @@ H5V_stride_optimize2(unsigned *np/*in,out*/, hsize_t *elmt_size/*in,out*/, break; case 3: /* For 2-D datasets */ - if((hsize_t)(stride1[2]) == *elmt_size && - (hsize_t)(stride2[2]) == *elmt_size) { + if(stride1[2] == *elmt_size && stride2[2] == *elmt_size) { *elmt_size *= size[2]; --*np; /* *np decrements to a value of 2 now */ - H5_CHECK_OVERFLOW(size[2],hsize_t,hssize_t); - stride1[1] += (hssize_t)size[2] * stride1[2]; - stride2[1] += (hssize_t)size[2] * stride2[2]; + stride1[1] += size[2] * stride1[2]; + stride2[1] += size[2] * stride2[2]; - if((hsize_t)(stride1[1]) == *elmt_size && - (hsize_t)(stride2[1]) == *elmt_size) { + if(stride1[1] == *elmt_size && stride2[1] == *elmt_size) { *elmt_size *= size[1]; --*np; /* *np decrements to a value of 1 now */ - H5_CHECK_OVERFLOW(size[1],hsize_t,hssize_t); - stride1[0] += (hssize_t)size[1] * stride1[1]; - stride2[0] += (hssize_t)size[1] * stride2[1]; + stride1[0] += size[1] * stride1[1]; + stride2[0] += size[1] * stride2[1]; - if((hsize_t)(stride1[0]) == *elmt_size && - (hsize_t)(stride2[0]) == *elmt_size) { + if(stride1[0] == *elmt_size && stride2[0] == *elmt_size) { *elmt_size *= size[0]; --*np; /* *np decrements to a value of 0 now */ } /* end if */ @@ -186,32 +175,25 @@ H5V_stride_optimize2(unsigned *np/*in,out*/, hsize_t *elmt_size/*in,out*/, break; case 4: /* For 3-D datasets */ - if((hsize_t)(stride1[3]) == *elmt_size && - (hsize_t)(stride2[3]) == *elmt_size) { + if(stride1[3] == *elmt_size && stride2[3] == *elmt_size) { *elmt_size *= size[3]; --*np; /* *np decrements to a value of 3 now */ - H5_CHECK_OVERFLOW(size[3],hsize_t,hssize_t); - stride1[2] += (hssize_t)size[3] * stride1[3]; - stride2[2] += (hssize_t)size[3] * stride2[3]; + stride1[2] += size[3] * stride1[3]; + stride2[2] += size[3] * stride2[3]; - if((hsize_t)(stride1[2]) == *elmt_size && - (hsize_t)(stride2[2]) == *elmt_size) { + if(stride1[2] == *elmt_size && stride2[2] == *elmt_size) { *elmt_size *= size[2]; --*np; /* *np decrements to a value of 2 now */ - H5_CHECK_OVERFLOW(size[2],hsize_t,hssize_t); - stride1[1] += (hssize_t)size[2] * stride1[2]; - stride2[1] += (hssize_t)size[2] * stride2[2]; + stride1[1] += size[2] * stride1[2]; + stride2[1] += size[2] * stride2[2]; - if((hsize_t)(stride1[1]) == *elmt_size && - (hsize_t)(stride2[1]) == *elmt_size) { + if(stride1[1] == *elmt_size && stride2[1] == *elmt_size) { *elmt_size *= size[1]; --*np; /* *np decrements to a value of 1 now */ - H5_CHECK_OVERFLOW(size[1],hsize_t,hssize_t); - stride1[0] += (hssize_t)size[1] * stride1[1]; - stride2[0] += (hssize_t)size[1] * stride2[1]; + stride1[0] += size[1] * stride1[1]; + stride2[0] += size[1] * stride2[1]; - if((hsize_t)(stride1[0]) == *elmt_size && - (hsize_t)(stride2[0]) == *elmt_size) { + if(stride1[0] == *elmt_size && stride2[0] == *elmt_size) { *elmt_size *= size[0]; --*np; /* *np decrements to a value of 0 now */ } /* end if */ @@ -222,13 +204,12 @@ H5V_stride_optimize2(unsigned *np/*in,out*/, hsize_t *elmt_size/*in,out*/, default: while (*np && - (hsize_t)(stride1[*np-1]) == *elmt_size && - (hsize_t)(stride2[*np-1]) == *elmt_size) { + stride1[*np-1] == *elmt_size && + stride2[*np-1] == *elmt_size) { *elmt_size *= size[*np-1]; if (--*np) { - H5_CHECK_OVERFLOW(size[*np],hsize_t,hssize_t); - stride1[*np-1] += (hssize_t)size[*np] * stride1[*np]; - stride2[*np-1] += (hssize_t)size[*np] * stride2[*np]; + stride1[*np-1] += size[*np] * stride1[*np]; + stride2[*np-1] += size[*np] * stride2[*np]; } } break; @@ -272,12 +253,11 @@ H5V_stride_optimize2(unsigned *np/*in,out*/, hsize_t *elmt_size/*in,out*/, */ hsize_t H5V_hyper_stride(unsigned n, const hsize_t *size, - const hsize_t *total_size, const hssize_t *offset, - hssize_t *stride/*out*/) + const hsize_t *total_size, const hsize_t *offset, + hsize_t *stride/*out*/) { hsize_t skip; /*starting point byte offset */ hsize_t acc; /*accumulator */ - hsize_t tmp; int i; /*counter */ hsize_t ret_value; /* Return value */ @@ -291,47 +271,41 @@ H5V_hyper_stride(unsigned n, const hsize_t *size, /* init */ assert(n>0); stride[n-1] = 1; - skip = offset ? (hsize_t)offset[n-1] : 0; + skip = offset ? offset[n-1] : 0; switch(n) { case 2: /* 1-D dataset */ - tmp = total_size[1] - size[1]; - assert (tmp<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - stride[0] = (hssize_t)tmp; /*overflow checked*/ + assert (total_size[1]>=size[1]); + stride[0] = total_size[1]-size[1]; /*overflow checked*/ acc = total_size[1]; - skip += acc * (offset ? (hsize_t)offset[0] : 0); + skip += acc * (offset ? offset[0] : 0); break; case 3: /* 2-D dataset */ - tmp = total_size[2] - size[2]; - assert (tmp<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - stride[1] = (hssize_t)tmp; /*overflow checked*/ + assert (total_size[2]>=size[2]); + stride[1] = total_size[2]-size[2]; /*overflow checked*/ acc = total_size[2]; skip += acc * (offset ? (hsize_t)offset[1] : 0); - tmp = acc * (total_size[1] - size[1]); - assert (tmp<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - stride[0] = (hssize_t)tmp; /*overflow checked*/ + assert (total_size[1]>=size[1]); + stride[0] = acc * (total_size[1] - size[1]); /*overflow checked*/ acc *= total_size[1]; skip += acc * (offset ? (hsize_t)offset[0] : 0); break; case 4: /* 3-D dataset */ - tmp = total_size[3] - size[3]; - assert (tmp<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - stride[2] = (hssize_t)tmp; /*overflow checked*/ + assert (total_size[3]>=size[3]); + stride[2] = total_size[3]-size[3]; /*overflow checked*/ acc = total_size[3]; skip += acc * (offset ? (hsize_t)offset[2] : 0); - tmp = acc * (total_size[2] - size[2]); - assert (tmp<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - stride[1] = (hssize_t)tmp; /*overflow checked*/ + assert (total_size[2]>=size[2]); + stride[1] = acc * (total_size[2] - size[2]); /*overflow checked*/ acc *= total_size[2]; skip += acc * (offset ? (hsize_t)offset[1] : 0); - tmp = acc * (total_size[1] - size[1]); - assert (tmp<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - stride[0] = (hssize_t)tmp; /*overflow checked*/ + assert (total_size[1]>=size[1]); + stride[0] = acc * (total_size[1] - size[1]); /*overflow checked*/ acc *= total_size[1]; skip += acc * (offset ? (hsize_t)offset[0] : 0); break; @@ -339,9 +313,8 @@ H5V_hyper_stride(unsigned n, const hsize_t *size, default: /* others */ for (i=(int)(n-2), acc=1; i>=0; --i) { - tmp = acc * (total_size[i+1] - size[i+1]); - assert (tmp<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - stride[i] = (hssize_t)tmp; /*overflow checked*/ + assert (total_size[i+1]>=size[i+1]); + stride[i] = acc * (total_size[i+1] - size[i+1]); /*overflow checked*/ acc *= total_size[i+1]; skip += acc * (offset ? (hsize_t)offset[i] : 0); } @@ -379,18 +352,18 @@ H5V_hyper_stride(unsigned n, const hsize_t *size, *------------------------------------------------------------------------- */ htri_t -H5V_hyper_eq(int n, - const hssize_t *offset1, const hsize_t *size1, - const hssize_t *offset2, const hsize_t *size2) +H5V_hyper_eq(unsigned n, + const hsize_t *offset1, const hsize_t *size1, + const hsize_t *offset2, const hsize_t *size2) { hsize_t nelmts1 = 1, nelmts2 = 1; - int i; + unsigned i; htri_t ret_value=TRUE; /* Return value */ /* Use FUNC_ENTER_NOAPI_NOINIT_NOFUNC here to avoid performance issues */ FUNC_ENTER_NOAPI_NOINIT_NOFUNC(H5V_hyper_eq) - if (n <= 0) HGOTO_DONE(TRUE) + if (n == 0) HGOTO_DONE(TRUE) for (i=0; i<n; i++) { if ((offset1 ? offset1[i] : 0) != (offset2 ? offset2[i] : 0)) @@ -428,8 +401,8 @@ done: */ htri_t H5V_hyper_disjointp(unsigned n, - const hssize_t *offset1, const size_t *size1, - const hssize_t *offset2, const size_t *size2) + const hsize_t *offset1, const size_t *size1, + const hsize_t *offset2, const size_t *size2) { unsigned u; htri_t ret_value=FALSE; /* Return value */ @@ -440,17 +413,15 @@ H5V_hyper_disjointp(unsigned n, if (!n || !size1 || !size2) HGOTO_DONE(TRUE) for (u=0; u<n; u++) { - assert (size1[u]<HSSIZET_MAX); - assert (size2[u]<HSSIZET_MAX); + assert (size1[u]<HSIZET_MAX); + assert (size2[u]<HSIZET_MAX); if (0==size1[u] || 0==size2[u]) HGOTO_DONE(TRUE) if (((offset1?offset1[u]:0) < (offset2?offset2[u]:0) && - ((offset1?offset1[u]:0) + (hssize_t)size1[u] <= - (offset2?offset2[u]:0))) || + ((offset1?offset1[u]:0) + size1[u] <= (offset2?offset2[u]:0))) || ((offset2?offset2[u]:0) < (offset1?offset1[u]:0) && - ((offset2?offset2[u]:0) + (hssize_t)size2[u] <= - (offset1?offset1[u]:0)))) + ((offset2?offset2[u]:0) + size2[u] <= (offset1?offset1[u]:0)))) HGOTO_DONE(TRUE) } @@ -482,12 +453,12 @@ done: */ herr_t H5V_hyper_fill(unsigned n, const hsize_t *_size, - const hsize_t *total_size, const hssize_t *offset, void *_dst, + const hsize_t *total_size, const hsize_t *offset, void *_dst, unsigned fill_value) { uint8_t *dst = (uint8_t*)_dst; /*cast for ptr arithmetic */ hsize_t size[H5V_HYPER_NDIMS]; /*a modifiable copy of _size */ - hssize_t dst_stride[H5V_HYPER_NDIMS]; /*destination stride info */ + hsize_t dst_stride[H5V_HYPER_NDIMS]; /*destination stride info */ hsize_t dst_start; /*byte offset to start of stride*/ hsize_t elmt_size = 1; /*bytes per element */ herr_t ret_value; /*function return status */ @@ -562,22 +533,20 @@ herr_t H5V_hyper_copy(unsigned n, const hsize_t *_size, /*destination*/ - const hsize_t *dst_size, const hssize_t *dst_offset, + const hsize_t *dst_size, const hsize_t *dst_offset, void *_dst, /*source*/ - const hsize_t *src_size, const hssize_t *src_offset, + const hsize_t *src_size, const hsize_t *src_offset, const void *_src) { const uint8_t *src = (const uint8_t*)_src; /*cast for ptr arithmtc */ uint8_t *dst = (uint8_t*) _dst; /*cast for ptr arithmtc */ hsize_t size[H5V_HYPER_NDIMS]; /*a modifiable _size */ - hssize_t src_stride[H5V_HYPER_NDIMS]; /*source stride info */ - hssize_t dst_stride[H5V_HYPER_NDIMS]; /*dest stride info */ + hsize_t src_stride[H5V_HYPER_NDIMS]; /*source stride info */ + hsize_t dst_stride[H5V_HYPER_NDIMS]; /*dest stride info */ hsize_t dst_start, src_start; /*offset to start at */ hsize_t elmt_size = 1; /*element size in bytes */ - hsize_t tmp1; - hsize_t tmp2; herr_t ret_value; /*return status */ #ifndef NDEBUG unsigned u; @@ -618,96 +587,82 @@ H5V_hyper_copy(unsigned n, const hsize_t *_size, assert(n>0); dst_stride[n-1] = 1; src_stride[n-1] = 1; - dst_start = dst_offset ? (hsize_t)dst_offset[n-1] : 0; - src_start = src_offset ? (hsize_t)src_offset[n-1] : 0; + dst_start = dst_offset ? dst_offset[n-1] : 0; + src_start = src_offset ? src_offset[n-1] : 0; /* Unroll loop for common cases */ switch(n) { case 2: - tmp1 = (dst_size[1] - size[1]); - tmp2 = (src_size[1] - size[1]); - assert (tmp1<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - assert (tmp2<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - dst_stride[0] = (hssize_t)tmp1; /*overflow checked*/ - src_stride[0] = (hssize_t)tmp2; /*overflow checked*/ + assert (dst_size[1]>=size[1]); + assert (src_size[1]>=size[1]); + dst_stride[0] = dst_size[1] - size[1]; /*overflow checked*/ + src_stride[0] = src_size[1] - size[1]; /*overflow checked*/ dst_acc = dst_size[1]; src_acc = src_size[1]; - dst_start += dst_acc * (dst_offset ? (hsize_t)dst_offset[0] : 0); - src_start += src_acc * (src_offset ? (hsize_t)src_offset[0] : 0); + dst_start += dst_acc * (dst_offset ? dst_offset[0] : 0); + src_start += src_acc * (src_offset ? src_offset[0] : 0); break; case 3: - tmp1 = (dst_size[2] - size[2]); - tmp2 = (src_size[2] - size[2]); - assert (tmp1<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - assert (tmp2<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - dst_stride[1] = (hssize_t)tmp1; /*overflow checked*/ - src_stride[1] = (hssize_t)tmp2; /*overflow checked*/ + assert (dst_size[2]>=size[2]); + assert (src_size[2]>=size[2]); + dst_stride[1] = dst_size[2] - size[2]; /*overflow checked*/ + src_stride[1] = src_size[2] - size[2]; /*overflow checked*/ dst_acc = dst_size[2]; src_acc = src_size[2]; - dst_start += dst_acc * (dst_offset ? (hsize_t)dst_offset[1] : 0); - src_start += src_acc * (src_offset ? (hsize_t)src_offset[1] : 0); - - tmp1 = dst_acc * (dst_size[1] - size[1]); - tmp2 = src_acc * (src_size[1] - size[1]); - assert (tmp1<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - assert (tmp2<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - dst_stride[0] = (hssize_t)tmp1; /*overflow checked*/ - src_stride[0] = (hssize_t)tmp2; /*overflow checked*/ + dst_start += dst_acc * (dst_offset ? dst_offset[1] : 0); + src_start += src_acc * (src_offset ? src_offset[1] : 0); + + assert (dst_size[1]>=size[1]); + assert (src_size[1]>=size[1]); + dst_stride[0] = dst_acc * (dst_size[1] - size[1]); /*overflow checked*/ + src_stride[0] = src_acc * (src_size[1] - size[1]); /*overflow checked*/ dst_acc *= dst_size[1]; src_acc *= src_size[1]; - dst_start += dst_acc * (dst_offset ? (hsize_t)dst_offset[0] : 0); - src_start += src_acc * (src_offset ? (hsize_t)src_offset[0] : 0); + dst_start += dst_acc * (dst_offset ? dst_offset[0] : 0); + src_start += src_acc * (src_offset ? src_offset[0] : 0); break; case 4: - tmp1 = (dst_size[3] - size[3]); - tmp2 = (src_size[3] - size[3]); - assert (tmp1<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - assert (tmp2<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - dst_stride[2] = (hssize_t)tmp1; /*overflow checked*/ - src_stride[2] = (hssize_t)tmp2; /*overflow checked*/ + assert (dst_size[3]>=size[3]); + assert (src_size[3]>=size[3]); + dst_stride[2] = dst_size[3] - size[3]; /*overflow checked*/ + src_stride[2] = src_size[3] - size[3]; /*overflow checked*/ dst_acc = dst_size[3]; src_acc = src_size[3]; - dst_start += dst_acc * (dst_offset ? (hsize_t)dst_offset[2] : 0); - src_start += src_acc * (src_offset ? (hsize_t)src_offset[2] : 0); - - tmp1 = dst_acc * (dst_size[2] - size[2]); - tmp2 = src_acc * (src_size[2] - size[2]); - assert (tmp1<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - assert (tmp2<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - dst_stride[1] = (hssize_t)tmp1; /*overflow checked*/ - src_stride[1] = (hssize_t)tmp2; /*overflow checked*/ + dst_start += dst_acc * (dst_offset ? dst_offset[2] : 0); + src_start += src_acc * (src_offset ? src_offset[2] : 0); + + assert (dst_size[2]>=size[2]); + assert (src_size[2]>=size[2]); + dst_stride[1] = dst_acc * (dst_size[2] - size[2]); /*overflow checked*/ + src_stride[1] = src_acc * (src_size[2] - size[2]); /*overflow checked*/ dst_acc *= dst_size[2]; src_acc *= src_size[2]; - dst_start += dst_acc * (dst_offset ? (hsize_t)dst_offset[1] : 0); - src_start += src_acc * (src_offset ? (hsize_t)src_offset[1] : 0); - - tmp1 = dst_acc * (dst_size[1] - size[1]); - tmp2 = src_acc * (src_size[1] - size[1]); - assert (tmp1<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - assert (tmp2<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - dst_stride[0] = (hssize_t)tmp1; /*overflow checked*/ - src_stride[0] = (hssize_t)tmp2; /*overflow checked*/ + dst_start += dst_acc * (dst_offset ? dst_offset[1] : 0); + src_start += src_acc * (src_offset ? src_offset[1] : 0); + + assert (dst_size[1]>=size[1]); + assert (src_size[1]>=size[1]); + dst_stride[0] = dst_acc * (dst_size[1] - size[1]); /*overflow checked*/ + src_stride[0] = src_acc * (src_size[1] - size[1]); /*overflow checked*/ dst_acc *= dst_size[1]; src_acc *= src_size[1]; - dst_start += dst_acc * (dst_offset ? (hsize_t)dst_offset[0] : 0); - src_start += src_acc * (src_offset ? (hsize_t)src_offset[0] : 0); + dst_start += dst_acc * (dst_offset ? dst_offset[0] : 0); + src_start += src_acc * (src_offset ? src_offset[0] : 0); break; default: /* others */ for (ii=(int)(n-2), dst_acc=1, src_acc=1; ii>=0; --ii) { - tmp1 = dst_acc * (dst_size[ii+1] - size[ii+1]); - tmp2 = src_acc * (src_size[ii+1] - size[ii+1]); - assert (tmp1<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - assert (tmp2<((hsize_t)1<<(8*sizeof(hssize_t)-1))); - dst_stride[ii] = (hssize_t)tmp1; /*overflow checked*/ - src_stride[ii] = (hssize_t)tmp2; /*overflow checked*/ + assert (dst_size[ii+1]>=size[ii+1]); + assert (src_size[ii+1]>=size[ii+1]); + dst_stride[ii] = dst_acc * (dst_size[ii+1] - size[ii+1]); /*overflow checked*/ + src_stride[ii] = src_acc * (src_size[ii+1] - size[ii+1]); /*overflow checked*/ dst_acc *= dst_size[ii+1]; src_acc *= src_size[ii+1]; - dst_start += dst_acc * (dst_offset ? (hsize_t)dst_offset[ii] : 0); - src_start += src_acc * (src_offset ? (hsize_t)src_offset[ii] : 0); + dst_start += dst_acc * (dst_offset ? dst_offset[ii] : 0); + src_start += src_acc * (src_offset ? src_offset[ii] : 0); } break; } /* end switch */ @@ -742,7 +697,7 @@ H5V_hyper_copy(unsigned n, const hsize_t *_size, */ herr_t H5V_stride_fill(unsigned n, hsize_t elmt_size, const hsize_t *size, - const hssize_t *stride, void *_dst, unsigned fill_value) + const hsize_t *stride, void *_dst, unsigned fill_value) { uint8_t *dst = (uint8_t*)_dst; /*cast for ptr arithmetic */ hsize_t idx[H5V_HYPER_NDIMS]; /*1-origin indices */ @@ -801,6 +756,74 @@ H5V_stride_fill(unsigned n, hsize_t elmt_size, const hsize_t *size, */ herr_t H5V_stride_copy(unsigned n, hsize_t elmt_size, const hsize_t *size, + const hsize_t *dst_stride, void *_dst, + const hsize_t *src_stride, const void *_src) +{ + uint8_t *dst = (uint8_t*)_dst; /*cast for ptr arithmetic*/ + const uint8_t *src = (const uint8_t*) _src; /*cast for ptr arithmetic*/ + hsize_t idx[H5V_HYPER_NDIMS]; /*1-origin indices */ + hsize_t nelmts; /*num elements to copy */ + hsize_t i; /*counter */ + int j; /*counters */ + hbool_t carry; /*carray for subtraction*/ + + FUNC_ENTER_NOAPI_NOFUNC(H5V_stride_copy) + assert (elmt_size<SIZET_MAX); + + if (n) { + H5V_vector_cpy(n, idx, size); + nelmts = H5V_vector_reduce_product(n, size); + for (i=0; i<nelmts; i++) { + + /* Copy an element */ + H5_CHECK_OVERFLOW(elmt_size,hsize_t,size_t); + HDmemcpy(dst, src, (size_t)elmt_size); /*lint !e671 The elmt_size will be OK */ + + /* Decrement indices and advance pointers */ + for (j=(int)(n-1), carry=TRUE; j>=0 && carry; --j) { + src += src_stride[j]; + dst += dst_stride[j]; + + if (--idx[j]) + carry = FALSE; + else { + assert(size); + idx[j] = size[j]; + } + } + } + } else { + H5_CHECK_OVERFLOW(elmt_size,hsize_t,size_t); + HDmemcpy (dst, src, (size_t)elmt_size); /*lint !e671 The elmt_size will be OK */ + } + + FUNC_LEAVE_NOAPI(SUCCEED) +} + + +/*------------------------------------------------------------------------- + * Function: H5V_stride_copy_s + * + * Purpose: Uses DST_STRIDE and SRC_STRIDE to advance through the arrays + * DST and SRC while copying bytes from SRC to DST. This + * function minimizes the number of calls to memcpy() by + * combining various strides, but it will never touch memory + * outside the hyperslab defined by the strides. + * + * Note: If the src_stride is all zero and elmt_size is one, then it's + * probably more efficient to use H5V_stride_fill() instead. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Robb Matzke + * Saturday, October 11, 1997 + * + * Modifications: + * + *------------------------------------------------------------------------- + */ +herr_t +H5V_stride_copy_s(unsigned n, hsize_t elmt_size, const hsize_t *size, const hssize_t *dst_stride, void *_dst, const hssize_t *src_stride, const void *_src) { @@ -812,7 +835,7 @@ H5V_stride_copy(unsigned n, hsize_t elmt_size, const hsize_t *size, int j; /*counters */ hbool_t carry; /*carray for subtraction*/ - FUNC_ENTER_NOAPI_NOFUNC(H5V_stride_copy) + FUNC_ENTER_NOAPI_NOFUNC(H5V_stride_copy_s) assert (elmt_size<SIZET_MAX); if (n) { @@ -869,12 +892,12 @@ H5V_stride_copy2(hsize_t nelmts, hsize_t elmt_size, /* destination */ unsigned dst_n, const hsize_t *dst_size, - const hssize_t *dst_stride, + const hsize_t *dst_stride, void *_dst, /* source */ unsigned src_n, const hsize_t *src_size, - const hssize_t *src_stride, + const hsize_t *src_stride, const void *_src) { uint8_t *dst = (uint8_t *) _dst; @@ -1046,9 +1069,8 @@ H5V_array_down(unsigned n, const hsize_t *total_size, hsize_t *down) *------------------------------------------------------------------------- */ hsize_t -H5V_array_offset_pre(unsigned n, const hsize_t *acc, const hssize_t *offset) +H5V_array_offset_pre(unsigned n, const hsize_t *acc, const hsize_t *offset) { - hsize_t skip; /*starting point byte offset */ int i; /*counter */ hsize_t ret_value; /* Return value */ @@ -1059,13 +1081,8 @@ H5V_array_offset_pre(unsigned n, const hsize_t *acc, const hssize_t *offset) assert(offset); /* Compute offset in array */ - for (i=(int)(n-1), skip=0; i>=0; --i) { - H5_CHECK_OVERFLOW(offset[i],hssize_t,hsize_t); - skip += acc[i] * (hsize_t)offset[i]; - } /* end for */ - - /* Set return value */ - ret_value=skip; + for (i=(int)(n-1), ret_value=0; i>=0; --i) + ret_value += acc[i] * offset[i]; FUNC_LEAVE_NOAPI(ret_value) } /* end H5V_array_offset_pre() */ @@ -1092,7 +1109,7 @@ H5V_array_offset_pre(unsigned n, const hsize_t *acc, const hssize_t *offset) *------------------------------------------------------------------------- */ hsize_t -H5V_array_offset(unsigned n, const hsize_t *total_size, const hssize_t *offset) +H5V_array_offset(unsigned n, const hsize_t *total_size, const hsize_t *offset) { hsize_t acc_arr[H5V_HYPER_NDIMS]; /* Accumulated size of down dimensions */ hsize_t ret_value; /* Return value */ @@ -1136,7 +1153,7 @@ done: *------------------------------------------------------------------------- */ herr_t -H5V_array_calc(hsize_t offset, unsigned n, const hsize_t *total_size, hssize_t *coords) +H5V_array_calc(hsize_t offset, unsigned n, const hsize_t *total_size, hsize_t *coords) { hsize_t idx[H5V_HYPER_NDIMS]; /* Size of each dimension in bytes */ hsize_t acc; /* Size accumulator */ @@ -1159,8 +1176,7 @@ H5V_array_calc(hsize_t offset, unsigned n, const hsize_t *total_size, hssize_t * /* Compute the coordinates from the offset */ for(u=0; u<n; u++) { - H5_CHECK_OVERFLOW(offset/idx[u],hsize_t,hssize_t); /*lint !e771 idx will always be initialized */ - coords[u]=(hssize_t)(offset/idx[u]); + coords[u]=offset/idx[u]; offset %= idx[u]; } /* end for */ @@ -1211,10 +1227,10 @@ H5V_array_calc(hsize_t offset, unsigned n, const hsize_t *total_size, hssize_t * *------------------------------------------------------------------------- */ herr_t -H5V_chunk_index(unsigned ndims, const hssize_t *coord, const size_t *chunk, +H5V_chunk_index(unsigned ndims, const hsize_t *coord, const size_t *chunk, const hsize_t *down_nchunks, hsize_t *chunk_idx) { - hssize_t scaled_coord[H5V_HYPER_NDIMS]; /* Scaled, coordinates, in terms of chunks */ + hsize_t scaled_coord[H5V_HYPER_NDIMS]; /* Scaled, coordinates, in terms of chunks */ unsigned u; /* Local index variable */ FUNC_ENTER_NOAPI_NOFUNC(H5V_chunk_index) @@ -1226,10 +1242,8 @@ H5V_chunk_index(unsigned ndims, const hssize_t *coord, const size_t *chunk, assert(chunk_idx); /* Compute the scaled coordinates for actual coordinates */ - for(u=0; u<ndims; u++) { - H5_CHECK_OVERFLOW(chunk[u],size_t,hssize_t); - scaled_coord[u]=coord[u]/(hssize_t)chunk[u]; - } /* end for */ + for(u=0; u<ndims; u++) + scaled_coord[u]=coord[u]/chunk[u]; /* Compute the chunk index */ *chunk_idx=H5V_array_offset_pre(ndims,down_nchunks,scaled_coord); /*lint !e772 scaled_coord will always be initialized */ @@ -1268,9 +1282,10 @@ H5V_memcpyvv(void *_dst, { unsigned char *dst; /* Destination buffer pointer */ const unsigned char *src; /* Source buffer pointer */ + size_t total_size=0; /* Total size of sequence in bytes */ size_t size; /* Size of sequence in bytes */ size_t u,v; /* Local index variables */ - ssize_t ret_value=0; /* Return value */ + ssize_t ret_value; /* Return value */ FUNC_ENTER_NOAPI_NOFUNC(H5V_memcpyvv) @@ -1316,13 +1331,16 @@ H5V_memcpyvv(void *_dst, u++; /* Increment number of bytes copied */ - ret_value+=(ssize_t)size; + total_size+=size; } /* end for */ /* Update current sequence vectors */ *dst_curr_seq=u; *src_curr_seq=v; + /* Set return value */ + H5_ASSIGN_OVERFLOW(ret_value,total_size,size_t,ssize_t); + FUNC_LEAVE_NOAPI(ret_value) } /* end H5V_memcpyvv() */ |