The following fixes PR60635 - we fail to update re-gimplified stmts
in gimple_regimplify_operands (or anywhere near its callers AFAIK).
It seems coverage is quite low as this must otherwise have been
noticed earlier ...
Bootstrapped on x86_64-unknown-linux-gnu, testing in progress.
Richard.
2014-03-25 Richard Biener <rguent...@suse.de>
PR middle-end/60635
* gimplify-me.c (gimple_regimplify_operands): Update the
re-gimplifed stmt.
* gfortran.dg/lto/pr60635_0.f90: New testcase.
* gfortran.dg/lto/pr60635_1.c: Likewise.
Index: gcc/gimplify-me.c
===================================================================
*** gcc/gimplify-me.c (revision 208789)
--- gcc/gimplify-me.c (working copy)
*************** gimple_regimplify_operands (gimple stmt,
*** 315,320 ****
--- 315,322 ----
gsi_insert_after (gsi_p, post_stmt, GSI_NEW_STMT);
pop_gimplify_context (NULL);
+
+ update_stmt (stmt);
}
Index: gcc/testsuite/gfortran.dg/lto/pr60635_1.c
===================================================================
*** gcc/testsuite/gfortran.dg/lto/pr60635_1.c (revision 0)
--- gcc/testsuite/gfortran.dg/lto/pr60635_1.c (working copy)
***************
*** 0 ****
--- 1,14 ----
+ #include <stdint.h>
+ #include <stdbool.h>
+
+ static bool littleendian=true;
+
+ uint16_t bigendc16(union{uint16_t * n;uint8_t* b;}x){
+
+ if (!littleendian) return *x.n;
+
+ uint16_t res = ((uint16_t)(x.b[1])<<0) |
+ ((uint16_t)(x.b[0])<<8);
+ return res;
+ }
+
Index: gcc/testsuite/gfortran.dg/lto/pr60635_0.f90
===================================================================
*** gcc/testsuite/gfortran.dg/lto/pr60635_0.f90 (revision 0)
--- gcc/testsuite/gfortran.dg/lto/pr60635_0.f90 (working copy)
***************
*** 0 ****
--- 1,16 ----
+ ! { dg-lto-do link }
+ program test
+ use iso_fortran_env
+
+ interface
+ integer(int16) function bigendc16(x) bind(C)
+ import
+ integer(int16), intent(in) :: x
+ end function
+ end interface
+
+ integer(int16) :: x16 = 12345
+ x16 = bigendc16(x16)
+ print *,x16
+ end program
+
