Open GoogleCodeExporter opened 9 years ago
In scale_neon64.cc
// TODO(Yang Zhang): Investigate less load instructions for
// the x/dx stepping
#define LOAD1_DATA32_LANE(vn, n) \
"lsr %5, %3, #16 \n" \
"add %6, %1, %5, lsl #2 \n" \
"add %3, %3, %4 \n" \
MEMACCESS(6) \
"ld1 {"#vn".s}["#n"], [%6] \n"
%6 and %1 are 64 bit pointers, but %5 and %3 are 32 bit int's.
I think %5 (tmp) should be made into int64.
Can this line:
lsr %5, %3, #16
widen from %3 32 bit to %5 64 bit?
Original comment by fbarch...@google.com
on 22 Apr 2015 at 1:17
Original issue reported on code.google.com by
fbarch...@google.com
on 22 Apr 2015 at 1:05