https://community.intel.com/t5/Intel-oneAPI-Math-Kernel-Library/modification-of-MKL-source-code/m-p/910214#M12086 If you compiled netlib DGEMM with a vectorizing compiler, it ought to perform well on problems which don't need cache blocking. By removing the if tests which skip loops involving 0. operations, you would give ifort -O3 more freedom to optimize.<BR /> Fri, 14 Sep 2007 13:13:45 GMT TimP 2007-09-14T13:13:45Z https://community.intel.com/t5/Intel-oneAPI-Math-Kernel-Library/modification-of-MKL-source-code/m-p/910213#M12085 hello,<BR />i would like to perform the operation A*D*B, where D is diagonal, and A,B are appropriate dimensions using BLAS. there are no routines specific to diagonal matrices, so i was able to make a one-line modification to the DGEMM code available on netlib to implement this. it runs much slower than the BLAS in MKL (of course). is the source code available to make such a modification?<BR /> Tue, 11 Sep 2007 23:29:38 GMT https://community.intel.com/t5/Intel-oneAPI-Math-Kernel-Library/modification-of-MKL-source-code/m-p/910213#M12085 xilin1 2007-09-11T23:29:38Z https://community.intel.com/t5/Intel-oneAPI-Math-Kernel-Library/modification-of-MKL-source-code/m-p/910214#M12086 If you compiled netlib DGEMM with a vectorizing compiler, it ought to perform well on problems which don't need cache blocking. By removing the if tests which skip loops involving 0. operations, you would give ifort -O3 more freedom to optimize.<BR /> Fri, 14 Sep 2007 13:13:45 GMT https://community.intel.com/t5/Intel-oneAPI-Math-Kernel-Library/modification-of-MKL-source-code/m-p/910214#M12086 TimP 2007-09-14T13:13:45Z