Blob Blame Raw
commit 5c00e04a1b61475a7f731f8cfede114201815e0a
Author: Mark Wielaard <mark@klomp.org>
Date:   Sun Dec 9 23:25:05 2018 +0100

    Implement ppc64 lxvb16x as 128-bit vector load with reversed double words.
    
    This makes it possible for memcheck to know which part of the 128bit
    vector is defined, even if the load is partly beyond an addressable block.
    
    Partially resolves bug 386945.

diff --git a/VEX/priv/guest_ppc_toIR.c b/VEX/priv/guest_ppc_toIR.c
index 7af4973..ec2f90a 100644
--- a/VEX/priv/guest_ppc_toIR.c
+++ b/VEX/priv/guest_ppc_toIR.c
@@ -20702,54 +20702,29 @@ dis_vx_load ( UInt theInstr )
    {
       DIP("lxvb16x %d,r%u,r%u\n", (UInt)XT, rA_addr, rB_addr);
 
-      IRTemp byte[16];
-      int i;
-      UInt ea_off = 0;
-      IRExpr* irx_addr;
-      IRTemp tmp_low[9];
-      IRTemp tmp_hi[9];
+      /* The result of lxvb16x should be the same on big and little
+         endian systems. We do a host load, then reverse the bytes in
+         the double words. If the host load was little endian we swap
+         them around again. */
 
-      tmp_low[0] = newTemp( Ity_I64 );
-      tmp_hi[0] = newTemp( Ity_I64 );
-      assign( tmp_low[0], mkU64( 0 ) );
-      assign( tmp_hi[0], mkU64( 0 ) );
-
-      for ( i = 0; i < 8; i++ ) {
-         byte[i] = newTemp( Ity_I64 );
-         tmp_low[i+1] = newTemp( Ity_I64 );
-
-         irx_addr = binop( mkSzOp( ty, Iop_Add8 ), mkexpr( EA ),
-                           ty == Ity_I64 ? mkU64( ea_off ) : mkU32( ea_off ) );
-         ea_off += 1;
-
-         assign( byte[i], binop( Iop_Shl64,
-                                 unop( Iop_8Uto64,
-                                       load( Ity_I8, irx_addr ) ),
-                                 mkU8( 8 * ( 7 - i ) ) ) );
+      IRTemp high = newTemp(Ity_I64);
+      IRTemp high_rev = newTemp(Ity_I64);
+      IRTemp low = newTemp(Ity_I64);
+      IRTemp low_rev = newTemp(Ity_I64);
 
-         assign( tmp_low[i+1],
-                 binop( Iop_Or64,
-                        mkexpr( byte[i] ), mkexpr( tmp_low[i] ) ) );
-      }
+      IRExpr *t128 = load( Ity_V128, mkexpr( EA ) );
 
-      for ( i = 0; i < 8; i++ ) {
-         byte[i + 8] = newTemp( Ity_I64 );
-         tmp_hi[i+1] = newTemp( Ity_I64 );
+      assign( high, unop(Iop_V128HIto64, t128) );
+      assign( high_rev, unop(Iop_Reverse8sIn64_x1, mkexpr(high)) );
+      assign( low, unop(Iop_V128to64, t128) );
+      assign( low_rev, unop(Iop_Reverse8sIn64_x1, mkexpr(low)) );
 
-         irx_addr = binop( mkSzOp( ty, Iop_Add8 ), mkexpr( EA ),
-                           ty == Ity_I64 ? mkU64( ea_off ) : mkU32( ea_off ) );
-         ea_off += 1;
+      if (host_endness == VexEndnessLE)
+         t128 = binop( Iop_64HLtoV128, mkexpr (low_rev), mkexpr (high_rev) );
+      else
+         t128 = binop( Iop_64HLtoV128, mkexpr (high_rev), mkexpr (low_rev) );
 
-         assign( byte[i+8], binop( Iop_Shl64,
-                                   unop( Iop_8Uto64,
-                                         load( Ity_I8, irx_addr ) ),
-                                   mkU8( 8 * ( 7 - i ) ) ) );
-         assign( tmp_hi[i+1], binop( Iop_Or64,
-                                     mkexpr( byte[i+8] ),
-                                     mkexpr( tmp_hi[i] ) ) );
-      }
-      putVSReg( XT, binop( Iop_64HLtoV128,
-                           mkexpr( tmp_low[8] ), mkexpr( tmp_hi[8] ) ) );
+      putVSReg( XT, t128 );
       break;
    }