------- Comment #3 from brian dot e dot bliss at intel dot com 2009-08-12
21:29 -------
It doesn't look like the problem is confined to Intel64 code. Here's the IPF
assembly for the GOMP_loop_dynamic_start call:
addp4 r15 = r14, r0
adds r14 = -24, r37
mov r39 = r16
;;
mov r40 = r15
addp4 r41 = 4294967294 - 0x100000000, r0
addl r42 = 1, r0
mov r43 = r14
adds r44 = -32, r37
br.call.sptk.many b0 = GOMP_loop_dynamic_start#
The "addp" instruction clears bits 32:60 of the result. "adds" (14-byte
sign-extended immediate form) should be used.
--
brian dot e dot bliss at intel dot com changed:
What |Removed |Added
----------------------------------------------------------------------------
Summary|Bad Intel64 codegen for |Bad 64-bit codegen for
|modified 39495 testcase |modified 39495 testcase
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40967
