The R Project SVN R

Rev

Show changed files | Details | Compare with Previous | Blame | RSS feed

Filtering Options

Rev Age Author Path Log message Diff
74461 3045 d 19 h pd /branches/R-3-5-branch/ Create R-3-5-branch  
74030 3127 d 7 h kalibera /trunk/src/main/ Fix return type in 74029.  
74029 3127 d 7 h kalibera /trunk/src/main/ Protect fixes.  
73160 3252 d 20 h maechler /trunk/ array(""[0], 1) now gives NA  
72383 3414 d 6 h morgan /trunk/src/main/ do_length() missing PROTECT()

- ans unprotected when coerceVector() called
 
72294 3435 d 1 h kalibera /trunk/ Fix complex number multiplication in matrix product to match
scalar/element-wise complex number multplication in complex.c (the
difference was in treatment of Inf/NaN in inputs).
 
72179 3449 d 7 h kalibera /trunk/ Do not check _OPENMP value as it is unreliable with some compilers.  
72174 3450 d 5 h kalibera /trunk/ Configure check for OpenMP4 SIMD reduction.  
72154 3454 d 5 h ripley /trunk/src/ add casts for gcc  
72153 3454 d 6 h ripley /trunk/src/main/ add some explicit casts  
72142 3456 d 5 h kalibera /trunk/ Allow selection of matprod implementations (default, internal, blas and
experimental default.simd). Check inputs for NaN/Inf also for symmetrical
case. The default is to bypass BLAS when input may have NaN/Inf.
 
72083 3462 d 2 h kalibera /trunk/ Bypass ZGEMM for inputs with NaN/Inf. Fix backup C code for complex matprod
so that it does not turn NaN into NA unnecessarily. Fix complex
(t)crossprod to work when Fortan and C disagree on complex numbers layout
(previously only cmatprod supported this).
 
72059 3465 d 3 h kalibera /trunk/src/main/ Bypass BLAS DGEMM/DGEMV in crossprod and tcrossprod when the input has NaN
or Inf. Previously this was done only in matprod. Fix mayHaveNaNOrInf for
long vectors.
 
72036 3468 d 6 h kalibera /trunk/ Use DGEMV instead of DGEMM for matrix-vector and vector-matrix
multiplication (performance optimization).
 
72020 3472 d 2 h kalibera /trunk/src/main/ Use simple C implementation instead of DGEMM in matprod also when the input
has Inf values (previously only NaN). Use a fast NaN/Inf checking loop
(from pqR with slight modification).
 
71956 3485 d 4 h ripley /trunk/ update dates  
71927 3488 d 1 h lawrence /trunk/src/main/ fix c71907  
71907 3490 d 17 h lawrence /trunk/src/main/ lengths() carries over dim and dimnames attributes  
70665 3715 d 23 h maechler /trunk/ diag(x, nr=3) now keeps typeof(x), also for logical, integer and raw  
70543 3745 d 19 h murdoch /trunk/ Missed type checking before using LENGTH(). (PR#16853)  

Show All