[Bug regression/37686] New: Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[Bug regression/37686] New: Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[luisgpm at linux dot vnet dot ibm dot com]

After this patch:

http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=140660

Building bzip2 with peak flags AND -mcpu=power4 for ppc (-O3 -mcpu=power4
-ffast-math -funroll-loops -ftree-loop-linear -fpeel-loops) started ICE'ing.

Here is the backtrace:

#0  0x103a2c3c in gimple_bb (g=0x4) at
/home/luis/src/gcc/HEAD/gcc/gimple.h:1071
#1  0x103a2b80 in gsi_start (seq=0xf76882b0) at
/home/luis/src/gcc/HEAD/gcc/gimple.h:4298
#2  0x103a5a48 in gsi_start_phis (bb=0xf7664b40) at
/home/luis/src/gcc/HEAD/gcc/gimple-iterator.c:770
#3  0x105e601c in verify_stmts () at
/home/luis/src/gcc/HEAD/gcc/tree-cfg.c:4177
#4  0x107e80cc in verify_ssa (check_modified_stmt=1 '\001') at
/home/luis/src/gcc/HEAD/gcc/tree-ssa.c:750
#5  0x10476c30 in execute_function_todo (data=0x4043) at
/home/luis/src/gcc/HEAD/gcc/passes.c:999
#6  0x1047657c in do_per_function (callback=0x10476898 <execute_function_todo>,
data=0x4043) at /home/luis/src/gcc/HEAD/gcc/passes.c:841
#7  0x10476d38 in execute_todo (flags=16451) at
/home/luis/src/gcc/HEAD/gcc/passes.c:1025
#8  0x10477c54 in execute_one_pass (pass=0x10eeabd4) at
/home/luis/src/gcc/HEAD/gcc/passes.c:1301
#9  0x10477ec4 in execute_pass_list (pass=0x10eeabd4) at
/home/luis/src/gcc/HEAD/gcc/passes.c:1327
#10 0x10477ef0 in execute_pass_list (pass=0x10eeaa9c) at
/home/luis/src/gcc/HEAD/gcc/passes.c:1328
#11 0x10477ef0 in execute_pass_list (pass=0x10eea494) at
/home/luis/src/gcc/HEAD/gcc/passes.c:1328
#12 0x106656a8 in tree_rest_of_compilation (fndecl=0xf7e74180) at
/home/luis/src/gcc/HEAD/gcc/tree-optimize.c:418
#13 0x10930d78 in cgraph_expand_function (node=0xf7e78300) at
/home/luis/src/gcc/HEAD/gcc/cgraphunit.c:1038
#14 0x10930fe0 in cgraph_expand_all_functions () at
/home/luis/src/gcc/HEAD/gcc/cgraphunit.c:1097
#15 0x109317a8 in cgraph_optimize () at
/home/luis/src/gcc/HEAD/gcc/cgraphunit.c:1302
#16 0x10033998 in c_write_global_declarations () at
/home/luis/src/gcc/HEAD/gcc/c-decl.c:8073
#17 0x105c9624 in compile_file () at /home/luis/src/gcc/HEAD/gcc/toplev.c:979
#18 0x105cc214 in do_compile () at /home/luis/src/gcc/HEAD/gcc/toplev.c:2190
#19 0x105cc2a8 in toplev_main (argc=30, argv=0xff880e44) at
/home/luis/src/gcc/HEAD/gcc/toplev.c:2222
#20 0x100f90fc in main (argc=30, argv=0xff880e44) at
/home/luis/src/gcc/HEAD/gcc/main.c:35

The failing line is "return g->gsbase.bb;", because "g" isn't really a valid
pointer:

(gdb) p g
$1 = (const_gimple) 0x4


-- 
           Summary: Building of CPU2000's bzip2 with peak flags with -
                    mcpu=power4 fails with an ICE.
           Product: gcc
           Version: 4.4.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: regression
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: luisgpm at linux dot vnet dot ibm dot com
 GCC build triplet: powerpc*-*-*
  GCC host triplet: powerpc*-*-*
GCC target triplet: powerpc*-*-*


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug regression/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[rguenth at gcc dot gnu dot org]

------- Comment #1 from rguenth at gcc dot gnu dot org  2008-10-01 10:54 -------
Preprocessed source?


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |WAITING
           Keywords|                            |ice-on-valid-code
            Summary|Building of CPU2000's bzip2 |[4.4 Regression] Building of
                   |with peak flags with -      |CPU2000's bzip2 with peak
                   |mcpu=power4 fails with an   |flags with -mcpu=power4
                   |ICE.                        |fails with an ICE.
   Target Milestone|---                         |4.4.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug regression/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[luisgpm at linux dot vnet dot ibm dot com]

------- Comment #2 from luisgpm at linux dot vnet dot ibm dot com  2008-10-01 13:10 -------
Created an attachment (id=16439)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=16439&action=view)
Preprocessed source for reduced bzip2.c

Preprocessed source.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug regression/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[luisgpm at linux dot vnet dot ibm dot com]

------- Comment #3 from luisgpm at linux dot vnet dot ibm dot com  2008-10-01 13:10 -------
Created an attachment (id=16440)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=16440&action=view)
Reduced source for bzip2.c

Source reduced with delta


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug regression/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[luisgpm at linux dot vnet dot ibm dot com]

------- Comment #4 from luisgpm at linux dot vnet dot ibm dot com  2008-10-01 13:13 -------
Created an attachment (id=16441)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=16441&action=view)
Reduced source for bzip2.c

Indented reduced source.


-- 

luisgpm at linux dot vnet dot ibm dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #16440|0                           |1
        is obsolete|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug regression/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[luisgpm at linux dot vnet dot ibm dot com]

------- Comment #5 from luisgpm at linux dot vnet dot ibm dot com  2008-10-01 13:19 -------
I'm still trying to minimize even further the source. Will attach when i have
something better.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug regression/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[rguenth at gcc dot gnu dot org]

------- Comment #6 from rguenth at gcc dot gnu dot org  2008-10-01 14:12 -------
Doesn't ICE for me with a cross from x86_64,

./cc1 -quiet bzip2-reduced.i -O3 -ffast-math -funroll-loops -ftree-loop-linear
-fpeel-loops -mcpu=power4 {,-m32,-m64}


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug regression/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[luisgpm at linux dot vnet dot ibm dot com]

------- Comment #7 from luisgpm at linux dot vnet dot ibm dot com  2008-10-01 17:44 -------
I still can ICE it with the same flags in a native system. Any other info you'd
like to have available?

I have a more reduced source, will post it soon.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug regression/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[pinskia at gcc dot gnu dot org]

------- Comment #8 from pinskia at gcc dot gnu dot org  2008-10-01 20:13 -------
I can reproduce this even with a stage1 compiler.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |UNCONFIRMED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug regression/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[pinskia at gcc dot gnu dot org]

------- Comment #10 from pinskia at gcc dot gnu dot org  2008-10-01 20:14 -------
This was with:
GNU C (GCC) version 4.4.0 20080929 (experimental) [trunk revision 140763]
(powerpc64-unknown-linux-gnu)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug regression/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[pinskia at gcc dot gnu dot org]

------- Comment #9 from pinskia at gcc dot gnu dot org  2008-10-01 20:13 -------
#0  0x10195c2c in single_exit (loop=0xf7b31930) at
/home/apinski/src/other/gcc/gcc/cfgloop.c:1618
#1  0x106a1208 in number_of_latch_executions (loop=0xf7b31930) at
/home/apinski/src/other/gcc/gcc/tree-scalar-evolution.c:2375
#2  0x1072bc2c in canonicalize_loop_induction_variables (loop=0xf7b31930,
create_iv=1 '\001', ul=UL_SINGLE_ITER, 
    try_eval=1 '\001') at
/home/apinski/src/other/gcc/gcc/tree-ssa-loop-ivcanon.c:277
#3  0x1072be84 in canonicalize_induction_variables () at
/home/apinski/src/other/gcc/gcc/tree-ssa-loop-ivcanon.c:330
#4  0x1075f740 in tree_ssa_loop_ivcanon () at
/home/apinski/src/other/gcc/gcc/tree-ssa-loop.c:378
#5  0x10475700 in execute_one_pass (pass=0x10ebb2c8) at
/home/apinski/src/other/gcc/gcc/passes.c:1279


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug regression/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[pinskia at gcc dot gnu dot org]

------- Comment #11 from pinskia at gcc dot gnu dot org  2008-10-01 20:23 -------
If I change the GC Parameters, I get a different backtrace:
#0  0x105c9abc in gsi_stmt (i=Cannot access memory at address 0x4
) at /home/apinski/src/other/gcc/gcc/gimple.h:4392
#1  0x105dd4fc in verify_stmts () at
/home/apinski/src/other/gcc/gcc/tree-cfg.c:4179
#2  0x107dec80 in verify_ssa (check_modified_stmt=1 '\001') at
/home/apinski/src/other/gcc/gcc/tree-ssa.c:750

So this looks GC related.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug tree-optimization/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[pinskia at gcc dot gnu dot org]

------- Comment #12 from pinskia at gcc dot gnu dot org  2008-10-01 21:55 -------
This is far as reduced testcase I could get, if I remove anything that is
unrelated the ICE goes away:
unsigned char inUse[256];
unsigned char len[6][258];
int code[6][258];
unsigned int crc32Table[256] = { };
  unsigned int getGlobalCRC (void) { }
  int bsLive;
void bsW (int n, unsigned int v) {
 while (bsLive >= 8) {}
 }
 void hbAssignCodes (int * code,         unsigned char * length, int minLen,
int maxLen, int alphaSize) {
   int n, vec, i;
   for (n = minLen;n <= maxLen;n++)
       for (i = 0; i < alphaSize;i++)
      code[i] = vec;
   }
  void sendMTFValues (void) {
   int v, t, i, j, gs, ge, totc, bt, bc, iter;
   int nSelectors, alphaSize, minLen, maxLen, selCtr;
   int nGroups, nBytes;
 {
    while (1)
  {
  break;
  }
       hbAssignCodes (&code[t][0], &len[t][0], minLen, maxLen, alphaSize);
     unsigned char inUse16[16];
     for (i = 0;i < 16;i++)
 if (inUse16[i])
  {
      for (j = 0;j < 16;j++)
   if (inUse[i * 16 + j])    { }
    }
   }
   for (i = 0; i < nSelectors;i++)     { }
   for (t = 0; t < nGroups;t++)
 {
       int curr = len[t][0];
       for (i = 0; i < alphaSize;i++)
          while (curr < len[t][i])     { }
     }
   while (1)
       for (i = gs; i <= ge;i++)  { }
 }

--- CUT ---
I had to use -m64 to get the ICE.

I was able to get the ICE with:
-mcpu=power4 -m64 -mtune=cell -O3 -ftree-loop-linear --param ggc-min-expand=0
--param ggc-min-heapsize=0 -funroll-loops -fpreprocessed 

Note -fpreprocessed is really needed, I don't know why though.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2008-10-01 21:55:56
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug tree-optimization/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[jakub at gcc dot gnu dot org]

------- Comment #13 from jakub at gcc dot gnu dot org  2008-10-01 22:20 -------
Thanks, finally #c12 is something I can reproduce in native stage1.  Will
investigate this morning.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug tree-optimization/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[jakub at gcc dot gnu dot org]

------- Comment #14 from jakub at gcc dot gnu dot org  2008-10-02 08:51 -------
This is a simple buffer overflow in lambda-code.c.
It is easily reproduceable everywhere with the #c12 testcase with just
-O3 -ftree-loop-linear -funroll-loops
if you add an assert:
--- lambda-code.c.jj 2008-09-05 12:56:32.000000000 +0200
+++ lambda-code.c 2008-10-02 10:39:59.000000000 +0200
@@ -2765,6 +2765,7 @@ av_for_af (tree access_fun, lambda_vecto
           return false;

         var = am_vector_index_for_loop (am, CHREC_VARIABLE (access_fun));
+        gcc_assert (var < AM_NB_COLUMNS (am));
         cy[var] = int_cst_value (right);

         if (TREE_CODE (left) == POLYNOMIAL_CHREC)

On this testcase, AM_PARAMETERS (am) == NULL and AM_NB_INDUCTION_VARS (am) ==
2,
so AM_NB_COLUMNS (am) == 3, yet the code overwrites cy[8].


-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dberlin at gcc dot gnu dot
                   |                            |org, spop at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug tree-optimization/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[jakub at gcc dot gnu dot org]

------- Comment #15 from jakub at gcc dot gnu dot org  2008-10-02 09:05 -------
Given that the cy vector is supposed to have 3 parts, the first for ivs, the
second for (optional) parameters and the last entry for constant column, and
av_for_af_base adds stuff to constant column and to the parameters, I'd expect
av_for_af means to write the ivs entries only, so the gcc_assert probably
should
even assert that var < AM_NB_INDUCTION_VARS (am).  access_fun is
{0B, +, 4}_9 and AM_LOOP_NEST_NUM (am) == 1, which is far more than the 2 ivs
we have allocated space for.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug tree-optimization/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[jakub at gcc dot gnu dot org]

------- Comment #16 from jakub at gcc dot gnu dot org  2008-10-02 15:03 -------
Wonder whether just am_vector_index_for_loop shouldn't be:
static inline int
am_vector_index_for_loop (struct access_matrix *access_matrix, int loop_num)
{
  int depth1 = loop_depth (get_loop (loop_num));
  int depth2 = loop_depth (get_loop (AM_LOOP_NEST_NUM (access_matrix)));
  gcc_assert (depth1 >= depth2
              && depth1 - depth2 <= AM_NB_INDUCTION_VARS (access_matrix));
  return depth1 - depth2;
}
Certainly subtracting two loop_num numbers doesn't make much sense to me.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug tree-optimization/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[jakub at gcc dot gnu dot org]

------- Comment #17 from jakub at gcc dot gnu dot org  2008-10-02 19:40 -------
That bootstrapped/regtested, unfortunately the <= in the assert is not right.
And when I've changed the <= in the assert to <, pr29581.f90 ICEs, it has
AM_NB_INDUCTION_VARS in one particular case set to 2, and AM_LOOP_NEST_NUM's
loop_depth is 1, yet the inline is called with loop_depth 3.  So I guess I'll
leave this to Sebastian or Jan Sjodin who added all this stuff this May.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug tree-optimization/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[spop at gcc dot gnu dot org]

------- Comment #18 from spop at gcc dot gnu dot org  2008-10-03 15:49 -------
Mine.

We should do exactly what graphite is doing: register loops belonging
to the loop nest in a VEC, then iterate over the loop nest to find the
index in the Access Matrix.


-- 

spop at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |spop at gcc dot gnu dot org
                   |dot org                     |
             Status|NEW                         |ASSIGNED
   Last reconfirmed|2008-10-01 21:55:56         |2008-10-03 15:49:06
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

Re: [Bug tree-optimization/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[Sebastian Pop]

[-- Attachment #1: Type: text/plain, Size: 98 bytes --]

Here is a patch that should fix this bug.  Can somebody test that it fixes it?

Thanks,
Sebastian

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 1238_pr37686.diff --]
[-- Type: text/x-diff; name=1238_pr37686.diff, Size: 4494 bytes --]

Index: tree-loop-linear.c
===================================================================
--- tree-loop-linear.c	(revision 140668)
+++ tree-loop-linear.c	(working copy)
@@ -333,11 +333,16 @@ linear_transform_loops (void)
       lambda_loopnest before, after;
       lambda_trans_matrix trans;
       struct obstack lambda_obstack;
+      struct loop *loop;
+      VEC(loop_p,heap) *nest = VEC_alloc (loop_p, heap, 3);
 
       depth = perfect_loop_nest_depth (loop_nest);
       if (depth == 0)
 	continue;
 
+      for (loop = loop_nest; loop; loop = loop->inner)
+	VEC_safe_push (loop_p, heap, nest, loop);
+
       gcc_obstack_init (&lambda_obstack);
       VEC_truncate (tree, oldivs, 0);
       VEC_truncate (tree, invariants, 0);
@@ -350,8 +355,7 @@ linear_transform_loops (void)
 	goto free_and_continue;
       
       lambda_collect_parameters (datarefs, &lambda_parameters);
-      if (!lambda_compute_access_matrices (datarefs, lambda_parameters,
-					   loop_nest->num))
+      if (!lambda_compute_access_matrices (datarefs, lambda_parameters, nest))
 	goto free_and_continue;
 
       if (dump_file && (dump_flags & TDF_DETAILS))
@@ -410,6 +414,7 @@ linear_transform_loops (void)
       obstack_free (&lambda_obstack, NULL);
       free_dependence_relations (dependence_relations);
       free_data_refs (datarefs);
+      VEC_free (loop_p, heap, nest);
     }
 
   for (i = 0; VEC_iterate (gimple, remove_ivs, i, oldiv_stmt); i++)
Index: tree-data-ref.h
===================================================================
--- tree-data-ref.h	(revision 140668)
+++ tree-data-ref.h	(working copy)
@@ -128,13 +128,13 @@ typedef struct scop *scop_p;
 */
 struct access_matrix
 {
-  int loop_nest_num;
+  VEC (loop_p, heap) *loop_nest;
   int nb_induction_vars;
   VEC (tree, heap) *parameters;
   VEC (lambda_vector, heap) *matrix;
 };
 
-#define AM_LOOP_NEST_NUM(M) (M)->loop_nest_num
+#define AM_LOOP_NEST(M) (M)->loop_nest
 #define AM_NB_INDUCTION_VARS(M) (M)->nb_induction_vars
 #define AM_PARAMETERS(M) (M)->parameters
 #define AM_MATRIX(M) (M)->matrix
@@ -149,8 +149,14 @@ struct access_matrix
 static inline int
 am_vector_index_for_loop (struct access_matrix *access_matrix, int loop_num)
 {
-  gcc_assert (loop_num >= AM_LOOP_NEST_NUM (access_matrix));
-  return loop_num - AM_LOOP_NEST_NUM (access_matrix);
+  int i;
+  loop_p l;
+
+  for (i = 0; VEC_iterate (loop_p, AM_LOOP_NEST (access_matrix), i, l); i++)
+    if (l->num == loop_num)
+      return i;
+
+  gcc_unreachable();
 }
 
 int access_matrix_get_index_for_parameter (tree, struct access_matrix *);
@@ -581,7 +587,7 @@ bool lambda_transform_legal_p (lambda_tr
 void lambda_collect_parameters (VEC (data_reference_p, heap) *,
 				VEC (tree, heap) **);
 bool lambda_compute_access_matrices (VEC (data_reference_p, heap) *,
-				     VEC (tree, heap) *, int);
+				     VEC (tree, heap) *, VEC (loop_p, heap) *);
 
 /* In tree-data-ref.c  */
 void split_constant_offset (tree , tree *, tree *);
Index: lambda-code.c
===================================================================
--- lambda-code.c	(revision 140668)
+++ lambda-code.c	(working copy)
@@ -2786,17 +2786,15 @@ av_for_af (tree access_fun, lambda_vecto
 
 static bool
 build_access_matrix (data_reference_p data_reference,
-		     VEC (tree, heap) *parameters, int loop_nest_num)
+		     VEC (tree, heap) *parameters, VEC (loop_p, heap) *nest)
 {
   struct access_matrix *am = GGC_NEW (struct access_matrix);
   unsigned i, ndim = DR_NUM_DIMENSIONS (data_reference);
-  struct loop *loop = gimple_bb (DR_STMT (data_reference))->loop_father;
-  struct loop *loop_nest = get_loop (loop_nest_num);
-  unsigned nivs = loop_depth (loop) - loop_depth (loop_nest) + 1;
+  unsigned nivs = VEC_length (loop_p, nest);
   unsigned lambda_nb_columns;
   lambda_vector_vec_p matrix;
 
-  AM_LOOP_NEST_NUM (am) = loop_nest_num;
+  AM_LOOP_NEST (am) = nest;
   AM_NB_INDUCTION_VARS (am) = nivs;
   AM_PARAMETERS (am) = parameters;
 
@@ -2824,13 +2822,13 @@ build_access_matrix (data_reference_p da
 bool
 lambda_compute_access_matrices (VEC (data_reference_p, heap) *datarefs,
 				VEC (tree, heap) *parameters,
-				int loop_nest_num)
+				VEC (loop_p, heap) *nest)
 {
   data_reference_p dataref;
   unsigned ix;
 
   for (ix = 0; VEC_iterate (data_reference_p, datarefs, ix, dataref); ix++)
-    if (!build_access_matrix (dataref, parameters, loop_nest_num))
+    if (!build_access_matrix (dataref, parameters, nest))
       return false;
 
   return true;

[Bug tree-optimization/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[sebpop at gmail dot com]

------- Comment #19 from sebpop at gmail dot com  2008-10-03 20:15 -------
Subject: Re:  [4.4 Regression] Building of CPU2000's bzip2 with peak flags with
-mcpu=power4 fails with an ICE.

Here is a patch that should fix this bug.  Can somebody test that it fixes it?

Thanks,
Sebastian


------- Comment #20 from sebpop at gmail dot com  2008-10-03 20:15 -------
Created an attachment (id=16459)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=16459&action=view)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug tree-optimization/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[luisgpm at linux dot vnet dot ibm dot com]

------- Comment #21 from luisgpm at linux dot vnet dot ibm dot com  2008-10-03 20:59 -------
It fixes the bzip2 ICE.

Thanks,
Luis


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug tree-optimization/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[spop at gcc dot gnu dot org]

------- Comment #22 from spop at gcc dot gnu dot org  2008-10-15 16:14 -------
Fixed.


-- 

spop at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686

[Bug tree-optimization/37686] [4.4 Regression] Building of CPU2000's bzip2 with peak flags with -mcpu=power4 fails with an ICE.

[spop at gcc dot gnu dot org]

------- Comment #23 from spop at gcc dot gnu dot org  2008-10-15 16:14 -------
Subject: Bug 37686

Author: spop
Date: Wed Oct 15 16:13:33 2008
New Revision: 141141

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=141141
Log:
2008-10-15  Sebastian Pop  <sebastian.pop@amd.com>

        PR tree-optimization/37686
        * testsuite/gcc.dg/tree-ssa/pr37686.c: New.
        * tree-loop-linear.c (linear_transform_loops): Build a
        loop nest vector.  Pass it to lambda_compute_access_matrices.
        * tree-data-ref.h (struct access_matrix): Store the loop nest
        relative to which it encodes the information.
        (AM_LOOP_NEST_NUM): Renamed AM_LOOP_NEST.
        (am_vector_index_for_loop): Reimplemented: iterate over the
        loop nest for finding the loop index in the access matrix.
        (lambda_compute_access_matrices): Update declaration.
        * lambda-code.c (build_access_matrix): Pass the loop nest and
        record it.
        (lambda_compute_access_matrices): Same.


Added:
    trunk/gcc/testsuite/gcc.dg/tree-ssa/pr37686.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/lambda-code.c
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-data-ref.h
    trunk/gcc/tree-loop-linear.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37686