OpenRISC 1000
ORPSoCv2 does not implement overflow for l.add #101
Closed jeremybennett opened this issue over 14 years ago
jeremybennett commented over 14 years ago
<p>
<tt>l.add</tt> is <em>signed</em> addition, which should set the CY (for 1's complement overflow) and OV (for 2's complement overflow) flags appropriately, and if appropriate trigger the overflow exception.
</p>
<p>
The following code fragment shows that the OV flag is never set.
</p>
<pre>
/* Jeremy stuff. First two small positive numbers */
l.addi r4,r0,1
l.addi r5,r0,2
l.mfspr r3,r0,SPR_SR
l.jal _report
l.nop
l.add r3,r5,r4
l.mfspr r6,r0,SPR_SR
l.jal _report
l.nop
l.add r3,r0,r6
l.jal _report
l.nop
/* Two small negative numbers */
l.movhi r4,hi(0xffffffff)
l.ori r4,r4,lo(0xffffffff)
l.movhi r5,hi(0xfffffffe)
l.ori r5,r5,lo(0xfffffffe)
l.mfspr r3,r0,SPR_SR
l.jal _report
l.nop
l.add r3,r5,r4
l.mfspr r6,r0,SPR_SR
l.jal _report
l.nop
l.add r3,r0,r6
l.jal _report
l.nop
/* Two large positive numbers. 2's complement should overflow, but not
1's complement. So the OV flag should be set, but not the CY flag.
*/
l.movhi r4,hi(0x40000000)
l.ori r4,r4,lo(0x40000000)
l.movhi r5,hi(0x40000000)
l.ori r5,r5,lo(0x40000000)
l.mfspr r3,r0,SPR_SR
l.jal _report
l.nop
l.add r3,r5,r4
l.mfspr r6,r0,SPR_SR
l.jal _report
l.nop
l.add r3,r0,r6
l.jal _report
l.nop
/* Two large negative numbers. 2's complement should overflow, and
also 1's complement, since these are also very big unsigned
numbers. So the OV flag and CY flags should both be set. */
l.movhi r4,hi(0xbfffffff)
l.ori r4,r4,lo(0xbfffffff)
l.movhi r5,hi(0xbfffffff)
l.ori r5,r5,lo(0xbfffffff)
l.mfspr r3,r0,SPR_SR
l.jal _report
l.nop
l.add r3,r5,r4
l.mfspr r6,r0,SPR_SR
l.jal _report
l.nop
l.add r3,r0,r6
l.jal _report
l.nop
l.jal _or32_exit
l.nop
l.nopjulius was assigned over 14 years ago
rherveille commented over 14 years ago
Hi Jeremy,
That is correct. Overflow is not implemented at all in the OR1200. I added this in the new code, but it will take a while before all of that is working.
Richard
jeremybennett commented over 14 years ago
<p>
Hi Richard
</p>
<p>
Thanks for the update. I look forward to seeing the new code.
</p>
<p>
For testing you may like to know that Or1ksim does implement overflow (and overflow exceptions in SR) in the current SVN tree. It will shortly have that implemented for all arithmetic operations, together with matching tests which should be reusable.
</p>
<p>
Jeremy
</p>
olof commented over 13 years ago
Has this been tested in hardware yet?
jeremybennett commented over 13 years ago
<p>
Transferred to OpenRISC bugzilla (<a href="http://bugzilla.opencores.org/show_bug.cgi?id=16">Bug 16</a>).
</p>
<p>
Marking closed in this bugtracker.
</p>
jeremybennett closed this over 13 years ago
© copyright 1999-2018 OpenCores.org, equivalent to Oliscience, all rights reserved. OpenCores®, registered trademark.