floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Make rpcc() on arm64 get closer to what x86 returns 

The Arm implementation of rpcc() uses the architected timer
which is defined by the SBSA to be between 10-400MHz. These numbers
are much smaller than the cycle counter frequency used by x86. Make
the numbers closer by shifting the cycle counter up by the number of
leading zeros in the cntfrq_el0 register which gets us closer to a
noraml cpu clock cycle range.
This commit is contained in:
Ali Saidi 2020-02-22 05:07:55 +00:00
parent 430ee31e66
commit 19f3a4091c
1 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
common_arm64.h
View File

@ -81,10 +81,12 @@ static void __inline blas_lock(volatile BLASULONG *address){
#if !defined(OS_DARWIN) && !defined (OS_ANDROID)
static __inline BLASULONG rpcc(void){
BLASULONG ret = 0;
blasint shift;
__asm__ __volatile__ ("isb; mrs %0,cntvct_el0":"=r"(ret));
__asm__ __volatile__ ("mrs %0,cntfrq_el0; clz %w0, %w0":"=&r"(shift));
return ret;
return ret << shift;
}
#define RPCC_DEFINED