|
 |
1b2d1f4 |
commit dc1523fb3550b4ed9dd4c178741626daaa474da7
|
|
|
1b2d1f4 |
Author: Mark Wielaard <mark@klomp.org>
|
|
|
1b2d1f4 |
Date: Mon Dec 10 17:18:20 2018 +0100
|
|
|
1b2d1f4 |
|
|
|
1b2d1f4 |
PR386945 set_AV_CR6 patch
|
|
|
1b2d1f4 |
|
|
|
1b2d1f4 |
https://bugs.kde.org/show_bug.cgi?id=386945#c62
|
|
|
1b2d1f4 |
|
|
|
1b2d1f4 |
diff --git a/VEX/priv/guest_ppc_toIR.c b/VEX/priv/guest_ppc_toIR.c
|
|
|
1b2d1f4 |
index ec2f90a..c3cc6d0 100644
|
|
|
1b2d1f4 |
--- a/VEX/priv/guest_ppc_toIR.c
|
|
|
1b2d1f4 |
+++ b/VEX/priv/guest_ppc_toIR.c
|
|
|
1b2d1f4 |
@@ -2062,45 +2062,88 @@ static void set_CR0 ( IRExpr* result )
|
|
|
1b2d1f4 |
static void set_AV_CR6 ( IRExpr* result, Bool test_all_ones )
|
|
|
1b2d1f4 |
{
|
|
|
1b2d1f4 |
/* CR6[0:3] = {all_ones, 0, all_zeros, 0}
|
|
|
1b2d1f4 |
- all_ones = (v[0] && v[1] && v[2] && v[3])
|
|
|
1b2d1f4 |
- all_zeros = ~(v[0] || v[1] || v[2] || v[3])
|
|
|
1b2d1f4 |
+ 32 bit: all_zeros = (v[0] || v[1] || v[2] || v[3]) == 0x0000'0000
|
|
|
1b2d1f4 |
+ all_ones = ~(v[0] && v[1] && v[2] && v[3]) == 0x0000'0000
|
|
|
1b2d1f4 |
+ where v[] denotes 32-bit lanes
|
|
|
1b2d1f4 |
+ or
|
|
|
1b2d1f4 |
+ 64 bit: all_zeros = (v[0] || v[1]) == 0x0000'0000'0000'0000
|
|
|
1b2d1f4 |
+ all_ones = ~(v[0] && v[1]) == 0x0000'0000'0000'0000
|
|
|
1b2d1f4 |
+ where v[] denotes 64-bit lanes
|
|
|
1b2d1f4 |
+
|
|
|
1b2d1f4 |
+ The 32- and 64-bit versions compute the same thing, but the 64-bit one
|
|
|
1b2d1f4 |
+ tries to be a bit more efficient.
|
|
|
1b2d1f4 |
*/
|
|
|
1b2d1f4 |
- IRTemp v0 = newTemp(Ity_V128);
|
|
|
1b2d1f4 |
- IRTemp v1 = newTemp(Ity_V128);
|
|
|
1b2d1f4 |
- IRTemp v2 = newTemp(Ity_V128);
|
|
|
1b2d1f4 |
- IRTemp v3 = newTemp(Ity_V128);
|
|
|
1b2d1f4 |
- IRTemp rOnes = newTemp(Ity_I8);
|
|
|
1b2d1f4 |
- IRTemp rZeros = newTemp(Ity_I8);
|
|
|
1b2d1f4 |
-
|
|
|
1b2d1f4 |
vassert(typeOfIRExpr(irsb->tyenv,result) == Ity_V128);
|
|
|
1b2d1f4 |
|
|
|
1b2d1f4 |
- assign( v0, result );
|
|
|
1b2d1f4 |
- assign( v1, binop(Iop_ShrV128, result, mkU8(32)) );
|
|
|
1b2d1f4 |
- assign( v2, binop(Iop_ShrV128, result, mkU8(64)) );
|
|
|
1b2d1f4 |
- assign( v3, binop(Iop_ShrV128, result, mkU8(96)) );
|
|
|
1b2d1f4 |
+ IRTemp overlappedOred = newTemp(Ity_V128);
|
|
|
1b2d1f4 |
+ IRTemp overlappedAnded = newTemp(Ity_V128);
|
|
|
1b2d1f4 |
+
|
|
|
1b2d1f4 |
+ if (mode64) {
|
|
|
1b2d1f4 |
+ IRTemp v0 = newTemp(Ity_V128);
|
|
|
1b2d1f4 |
+ IRTemp v1 = newTemp(Ity_V128);
|
|
|
1b2d1f4 |
+ assign( v0, result );
|
|
|
1b2d1f4 |
+ assign( v1, binop(Iop_ShrV128, result, mkU8(64)) );
|
|
|
1b2d1f4 |
+ assign(overlappedOred,
|
|
|
1b2d1f4 |
+ binop(Iop_OrV128, mkexpr(v0), mkexpr(v1)));
|
|
|
1b2d1f4 |
+ assign(overlappedAnded,
|
|
|
1b2d1f4 |
+ binop(Iop_AndV128, mkexpr(v0), mkexpr(v1)));
|
|
|
1b2d1f4 |
+ } else {
|
|
|
1b2d1f4 |
+ IRTemp v0 = newTemp(Ity_V128);
|
|
|
1b2d1f4 |
+ IRTemp v1 = newTemp(Ity_V128);
|
|
|
1b2d1f4 |
+ IRTemp v2 = newTemp(Ity_V128);
|
|
|
1b2d1f4 |
+ IRTemp v3 = newTemp(Ity_V128);
|
|
|
1b2d1f4 |
+ assign( v0, result );
|
|
|
1b2d1f4 |
+ assign( v1, binop(Iop_ShrV128, result, mkU8(32)) );
|
|
|
1b2d1f4 |
+ assign( v2, binop(Iop_ShrV128, result, mkU8(64)) );
|
|
|
1b2d1f4 |
+ assign( v3, binop(Iop_ShrV128, result, mkU8(96)) );
|
|
|
1b2d1f4 |
+ assign(overlappedOred,
|
|
|
1b2d1f4 |
+ binop(Iop_OrV128,
|
|
|
1b2d1f4 |
+ binop(Iop_OrV128, mkexpr(v0), mkexpr(v1)),
|
|
|
1b2d1f4 |
+ binop(Iop_OrV128, mkexpr(v2), mkexpr(v3))));
|
|
|
1b2d1f4 |
+ assign(overlappedAnded,
|
|
|
1b2d1f4 |
+ binop(Iop_AndV128,
|
|
|
1b2d1f4 |
+ binop(Iop_AndV128, mkexpr(v0), mkexpr(v1)),
|
|
|
1b2d1f4 |
+ binop(Iop_AndV128, mkexpr(v2), mkexpr(v3))));
|
|
|
1b2d1f4 |
+ }
|
|
|
1b2d1f4 |
+
|
|
|
1b2d1f4 |
+ IRTemp rOnes = newTemp(Ity_I8);
|
|
|
1b2d1f4 |
+ IRTemp rZeroes = newTemp(Ity_I8);
|
|
|
1b2d1f4 |
|
|
|
1b2d1f4 |
- assign( rZeros, unop(Iop_1Uto8,
|
|
|
1b2d1f4 |
- binop(Iop_CmpEQ32, mkU32(0xFFFFFFFF),
|
|
|
1b2d1f4 |
- unop(Iop_Not32,
|
|
|
1b2d1f4 |
- unop(Iop_V128to32,
|
|
|
1b2d1f4 |
- binop(Iop_OrV128,
|
|
|
1b2d1f4 |
- binop(Iop_OrV128, mkexpr(v0), mkexpr(v1)),
|
|
|
1b2d1f4 |
- binop(Iop_OrV128, mkexpr(v2), mkexpr(v3))))
|
|
|
1b2d1f4 |
- ))) );
|
|
|
1b2d1f4 |
+ if (mode64) {
|
|
|
1b2d1f4 |
+ assign(rZeroes,
|
|
|
1b2d1f4 |
+ unop(Iop_1Uto8,
|
|
|
1b2d1f4 |
+ binop(Iop_CmpEQ64,
|
|
|
1b2d1f4 |
+ mkU64(0),
|
|
|
1b2d1f4 |
+ unop(Iop_V128to64, mkexpr(overlappedOred)))));
|
|
|
1b2d1f4 |
+ assign(rOnes,
|
|
|
1b2d1f4 |
+ unop(Iop_1Uto8,
|
|
|
1b2d1f4 |
+ binop(Iop_CmpEQ64,
|
|
|
1b2d1f4 |
+ mkU64(0),
|
|
|
1b2d1f4 |
+ unop(Iop_Not64,
|
|
|
1b2d1f4 |
+ unop(Iop_V128to64, mkexpr(overlappedAnded))))));
|
|
|
1b2d1f4 |
+ } else {
|
|
|
1b2d1f4 |
+ assign(rZeroes,
|
|
|
1b2d1f4 |
+ unop(Iop_1Uto8,
|
|
|
1b2d1f4 |
+ binop(Iop_CmpEQ32,
|
|
|
1b2d1f4 |
+ mkU32(0),
|
|
|
1b2d1f4 |
+ unop(Iop_V128to32, mkexpr(overlappedOred)))));
|
|
|
1b2d1f4 |
+ assign(rOnes,
|
|
|
1b2d1f4 |
+ unop(Iop_1Uto8,
|
|
|
1b2d1f4 |
+ binop(Iop_CmpEQ32,
|
|
|
1b2d1f4 |
+ mkU32(0),
|
|
|
1b2d1f4 |
+ unop(Iop_Not32,
|
|
|
1b2d1f4 |
+ unop(Iop_V128to32, mkexpr(overlappedAnded))))));
|
|
|
1b2d1f4 |
+ }
|
|
|
1b2d1f4 |
+
|
|
|
1b2d1f4 |
+ // rOnes might not be used below. But iropt will remove it, so there's no
|
|
|
1b2d1f4 |
+ // inefficiency as a result.
|
|
|
1b2d1f4 |
|
|
|
1b2d1f4 |
if (test_all_ones) {
|
|
|
1b2d1f4 |
- assign( rOnes, unop(Iop_1Uto8,
|
|
|
1b2d1f4 |
- binop(Iop_CmpEQ32, mkU32(0xFFFFFFFF),
|
|
|
1b2d1f4 |
- unop(Iop_V128to32,
|
|
|
1b2d1f4 |
- binop(Iop_AndV128,
|
|
|
1b2d1f4 |
- binop(Iop_AndV128, mkexpr(v0), mkexpr(v1)),
|
|
|
1b2d1f4 |
- binop(Iop_AndV128, mkexpr(v2), mkexpr(v3)))
|
|
|
1b2d1f4 |
- ))) );
|
|
|
1b2d1f4 |
putCR321( 6, binop(Iop_Or8,
|
|
|
1b2d1f4 |
binop(Iop_Shl8, mkexpr(rOnes), mkU8(3)),
|
|
|
1b2d1f4 |
- binop(Iop_Shl8, mkexpr(rZeros), mkU8(1))) );
|
|
|
1b2d1f4 |
+ binop(Iop_Shl8, mkexpr(rZeroes), mkU8(1))) );
|
|
|
1b2d1f4 |
} else {
|
|
|
1b2d1f4 |
- putCR321( 6, binop(Iop_Shl8, mkexpr(rZeros), mkU8(1)) );
|
|
|
1b2d1f4 |
+ putCR321( 6, binop(Iop_Shl8, mkexpr(rZeroes), mkU8(1)) );
|
|
|
1b2d1f4 |
}
|
|
|
1b2d1f4 |
putCR0( 6, mkU8(0) );
|
|
|
1b2d1f4 |
}
|
|
|
1b2d1f4 |
diff --git a/memcheck/mc_translate.c b/memcheck/mc_translate.c
|
|
|
1b2d1f4 |
index c24db91..7f69ee3 100644
|
|
|
1b2d1f4 |
--- a/memcheck/mc_translate.c
|
|
|
1b2d1f4 |
+++ b/memcheck/mc_translate.c
|
|
|
1b2d1f4 |
@@ -8322,6 +8322,9 @@ IRSB* MC_(instrument) ( VgCallbackClosure* closure,
|
|
|
1b2d1f4 |
# elif defined(VGA_amd64)
|
|
|
1b2d1f4 |
mce.dlbo.dl_Add64 = DLauto;
|
|
|
1b2d1f4 |
mce.dlbo.dl_CmpEQ32_CmpNE32 = DLexpensive;
|
|
|
1b2d1f4 |
+# elif defined(VGA_ppc64le)
|
|
|
1b2d1f4 |
+ // Needed by (at least) set_AV_CR6() in the front end.
|
|
|
1b2d1f4 |
+ mce.dlbo.dl_CmpEQ64_CmpNE64 = DLexpensive;
|
|
|
1b2d1f4 |
# endif
|
|
|
1b2d1f4 |
|
|
|
1b2d1f4 |
/* preInstrumentationAnalysis() will allocate &mce.tmpHowUsed and then
|