Important Notice: Our web hosting provider recently started charging us for additional visits, which was unexpected. In response, we're seeking donations. Depending on the situation, we may explore different monetization options for our Community and Expert Contributors. It's crucial to provide more returns for their expertise and offer more Expert Validated Answers or AI Validated Answers. Learn more about our hosting issue here.

When using PETSc in single precision mode (–with-precision=single when running config/configure.py) are the operations done in single or double precision?

April 26, 2017config double mode operations petsc precision running with-precision=single

0

Posted

When using PETSc in single precision mode (–with-precision=single when running config/configure.py) are the operations done in single or double precision?

1 Answer

0

Posted

PETSc does NOT do any explicit conversion of single precision to double before performing computations; this it depends on the hardware and compiler what happens. For example, the compiler could choose to put the single precision numbers into the usual double precision registers and then use the usual double precision floating point unit. Or it could use SSE2 instructions that work directly on the single precision numbers. It is a bit of a mystery what decisions get made sometimes. There may be compiler flags in some circumstances that can affect this.