[llvm-commits] [llvm] r112206 - /llvm/trunk/lib/Target/ARM/ARMInstrInfo.td

Bill Wendling isanbard at gmail.com
Thu Aug 26 11:33:51 PDT 2010


Author: void
Date: Thu Aug 26 13:33:51 2010
New Revision: 112206

URL: http://llvm.org/viewvc/llvm-project?rev=112206&view=rev
Log:
Reapply r112176 without removing the other CMN patterns (that was unintentional).

Modified:
    llvm/trunk/lib/Target/ARM/ARMInstrInfo.td

Modified: llvm/trunk/lib/Target/ARM/ARMInstrInfo.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMInstrInfo.td?rev=112206&r1=112205&r2=112206&view=diff
==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMInstrInfo.td (original)
+++ llvm/trunk/lib/Target/ARM/ARMInstrInfo.td Thu Aug 26 13:33:51 2010
@@ -2297,8 +2297,52 @@
 
 defm CMP  : AI1_cmp_irs<0b1010, "cmp",
                         BinOpFrag<(ARMcmp node:$LHS, node:$RHS)>>;
-//FIXME: Disable CMN, as CCodes are backwards from compare expectations
-//       Compare-to-zero still works out, just not the relationals
+
+// FIXME: There seems to be a (potential) hardware bug with the CMN instruction
+// and comparison with 0. These two pieces of code should give identical
+// results:
+//
+//   rsbs r1, r1, 0
+//   cmp  r0, r1
+//   mov  r0, #0
+//   it   ls
+//   mov  r0, #1
+//
+// and:
+// 
+//   cmn  r0, r1
+//   mov  r0, #0
+//   it   ls
+//   mov  r0, #1
+//
+// However, the CMN gives the *opposite* result when r1 is 0. This is because
+// the carry flag is set in the CMP case but not in the CMN case. In short, the
+// CMP instruction doesn't perform a truncate of the (logical) NOT of 0 plus the
+// value of r0 and the carry bit (because the "carry bit" parameter to
+// AddWithCarry is defined as 1 in this case, the carry flag will always be set
+// when r0 >= 0). The CMN instruction doesn't perform a NOT of 0 so there is
+// never a "carry" when this AddWithCarry is performed (because the "carry bit"
+// parameter to AddWithCarry is defined as 0).
+//
+// The AddWithCarry in the CMP case seems to be relying upon the identity:
+// 
+//   ~x + 1 = -x
+//
+// However when x is 0 and unsigned, this doesn't hold:
+//
+//    x = 0
+//   ~x = 0xFFFF FFFF
+//   ~x + 1 = 0x1 0000 0000
+//   (-x = 0) != (0x1 0000 0000 = ~x + 1)
+//
+// Therefore, we should disable *all* versions of CMN, especially when comparing
+// against zero, until we can limit when the CMN instruction is used (when we
+// know that the RHS is not 0) or when we have a hardware fix for this.
+//
+// (See the ARM docs for the "AddWithCarry" pseudo-code.)
+//
+// This is related to <rdar://problem/7569620>.
+//
 //defm CMN  : AI1_cmp_irs<0b1011, "cmn",
 //                        BinOpFrag<(ARMcmp node:$LHS,(ineg node:$RHS))>>;
 





More information about the llvm-commits mailing list