search for: multiplicand

Fold MAD dst, src0, immed, src2 (or src0/immed swapped) when - immed = 0 -> MOV dst, src2 - immed = +/- 1 -> ADD dst, src0, src2 These types of MAD pattersn were observed in some st/nine shaders. Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu> --- Haven't tested this enough to push yet, but thought I'd get it out there. Passes some simple test cases.

...eers, Lang. On Sat, Dec 21, 2013 at 11:42 PM, Demikhovsky, Elena <elena.demikhovsky at intel.com> wrote: > 1) Our architects say that vmovaps %reg, %reg causes only register renaming in HW and does not affect performance. > > 2) The non-numeric result of FMA is not commutative for multiplicand and the multiplier. You are right. This property should be removed. > > > - Elena > > -----Original Message----- > From: Lang Hames [mailto:lhames at gmail.com] > Sent: Friday, December 20, 2013 23:03 > To: Kay Tiong Khoo > Cc: LLVM Developers Mailing List; Demikhovsky,...

Hi Kay, My patch will partially address your bug. For now I'm just looking to switch the default FMA from vfmadd213xx to vfmadd231xx. That will cause the code in PR17229 to compile as desired, but would regress code like: double foo(double a, double b, double c) { return a * b + c; } Which will now require a vmovaps + vfmadd231. If this impacts real benchmarks we could add an

This patch supplies software IEEE floating point support. The comment from the patch reproduced below says all there is to say. This patch contains the prior "cleanup" patch; please don't apply that one. Please let me know of any bugs. It is tested reasonably well, but until I put together random tests it's hard to have 100% confidence. Neil. /* A self-contained host- and

This is my 2007 New Year wishlist for R features: 1. Matrix Multiplication Enhance matrix multiplication to work with multidimensional arrays such that the last dimension of the first multiplicand must equal the first dimension of the second. See: https://www.stat.math.ethz.ch/pipermail/r-devel/2006-July/038497.html 2. Grid - logical-valued function as first arg of grid.edit - transparency under Windows (not sure if this involves grid or just the Windows graphics device)...

...antWidth++; // now process acording to type // AQUI AQUI - fazer ou achar a tabela abaixo if( cb->mapType ) { cb->quantNum = cb->codeEntries^(1/cb->dimensions); } else { cb->quantNum = cb->dimensions*cb->codeEntries; } // create multiplicands array and read them cb->quantValues = new byte[cb->quantNum]; if( cb->quantValues==NULL ) return 0; // read all quad values for(i=0; i<cb->quantNum; i++) this->fReadBits1((int*) &cb->quantValues[i], 5); break; default : return 0; } return...

Dear list, n00b question, but still can't find any easy answer. Here is a df: > df<-data.frame(cbind(x=c("AA","BB","CC","AA"),y=1:4)) > df x y 1 AA 1 2 BB 2 3 CC 3 4 AA 4 I want to modify this df this way : if df$x=="AA" then df$y=df$y*10 if df$x=="BB" then df$y=df$y*25 and so on with other conditions. TY for any

...= 4 - 2x - 2y + xy + * + * For signed multiplies, we subtract (x << 32) from the partial + * product to fix this problem for negative multipliers (see mul.s). + * Because of the way the shift into the partial product is calculated + * (N xor V), this term is automatically removed for the multiplicand, + * so we don't have to adjust. + * + * But for unsigned multiplies, the high order bit wasn't a sign bit, + * and the correction is wrong. So for unsigned multiplies where the + * high order bit is one, we end up with xy - (y << 32). To fix it + * we add y << 32. + */...

I have updated the klibc patchset based on feedback received. In particular, the patchset has been reorganized so as not to break git-bisect. Additionally, this updates the patch base to 2.6.17-git12 (d38b69689c349f35502b92e20dafb30c62d49d63) and klibc 1.4.8; the main difference on the klibc side is removal of obsolete code. This is also available as a git tree at: