Ramana Radhakrishnan wrote:
> PR51819 is a case where we were actually putting out alignment hints
> for the wrong memory size. The attached patch corrects this and
> another latent issue that I spotted.
Your patch did:
/* Only certain alignment specifiers are supported by the hardware. */
if (memsize == 16 && (align % 32) == 0)
align_bits = 256;
- else if ((memsize == 8 || memsize == 16) && (align % 16) == 0)
+ else if (memsize == 16 && (align % 16) == 0)
align_bits = 128;
- else if ((align % 8) == 0)
+ else if (memsize >= 8 && (align % 8) == 0)
align_bits = 64;
else
align_bits = 0;
However, I'm still seeing incorrect memory hints (with an add-on patch):
Error: bad alignment -- `vld1.64 {d18-d19},[r3:256]'
The ISA document specifies the following supported alignment hints:
64 8-byte alignment
128 16-byte alignment, available only if <list> contains two or four registers
256 32-byte alignment, available only if <list> contains four registers
Shouldn't the check be implemented along the following lines?
if (memsize == 32 && (align % 32) == 0)
align_bits = 256;
else if ((memsize == 16 || memsize == 32) && (align % 16) == 0)
align_bits = 128;
else if (memsize >= 8 && (align % 8) == 0)
align_bits = 64;
else
align_bits = 0;
Bye,
Ulrich
--
Dr. Ulrich Weigand
GNU Toolchain for Linux on System z and Cell BE
ulrich.weig...@de.ibm.com
