This problem appears when two packed 32-bit values are converted to two
floats. When doing sign extension of the top value, the 32-bit value is
shifted down before bit 63 is checked, thus packed integers in the top of
a 64-bit value aren't detected as negative properly.
This patch fixes things, but there must be a better way of doing it
without introducing a new variable.
This problem appears with the qemu test-x86_64 program.
Vince
diff -r a2a79fe9655d src/arch/x86/isa/microops/mediaop.isa
--- a/src/arch/x86/isa/microops/mediaop.isa Wed Nov 11 17:49:09 2009 -0500
+++ b/src/arch/x86/isa/microops/mediaop.isa Fri Nov 13 21:31:12 2009 -0500
@@ -1288,9 +1288,10 @@
for (int i = 0; i < items; i++) {
int srcHiIndex = srcStart + (i + 1) * srcSizeBits - 1;
int srcLoIndex = srcStart + (i + 0) * srcSizeBits;
+ int srcSignIndex = ((srcHiIndex+1)-srcLoIndex)-1;
uint64_t argBits = bits(FpSrcReg1.uqw, srcHiIndex, srcLoIndex);
- int64_t sArg = argBits | (0 - (argBits & (ULL(1) <<
srcHiIndex)));
+ int64_t sArg = argBits | (0 - (argBits & (ULL(1) <<
srcSignIndex)));
double arg = sArg;
if (destSize == 4) {
_______________________________________________
m5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/m5-dev
