Bug 33219

I did a little debugging on this, and I suspect the problem was added when the copy propagation pass was added.  The acp buffer is being over-run.  It seems the expectation that this->next_temp is greater than any in-use register index is not being held.  The assertion below fails for this test case.

#3  0x00007ffff5599bb2 in ir_to_mesa_visitor::copy_propagate (this=0x7fffffffdcb0) at program/ir_to_mesa.cpp:2657
(gdb) print inst->dst_reg.index
$3 = 2
(gdb) print this->next_temp
$4 = 2

As a result, the buffer gets over-run at either line 2729 or 2746.

This probably wouldn't crash without dynamic memory allocation 
diff --git a/src/mesa/program/ir_to_mesa.cpp b/src/mesa/program/ir_to_mesa.cpp
index 786fdfb..ddcf883 100644
--- a/src/mesa/program/ir_to_mesa.cpp
+++ b/src/mesa/program/ir_to_mesa.cpp
@@ -2652,6 +2652,8 @@ ir_to_mesa_visitor::copy_propagate(void)
 
    foreach_iter(exec_list_iterator, iter, this->instructions) {
       ir_to_mesa_instruction *inst = (ir_to_mesa_instruction *)iter.get();
+ 
+      assert(inst->dst_reg.index < this->next_temp);
 
       /* First, do any copy propagation possible into the src regs. */
       for (int r = 0; r < 3; r++) {

With a little more digging, it seems that there is still a variable in the linked program with an unset array size.  This causes type_size to return 0, and next_temp does not get advanced.  This should not be able to happen due to checks in the linker for 0-sized arrays.

diff --git a/src/mesa/program/ir_to_mesa.cpp b/src/mesa/program/ir_to_mesa.cpp
index 786fdfb..7b5b424 100644
--- a/src/mesa/program/ir_to_mesa.cpp
+++ b/src/mesa/program/ir_to_mesa.cpp
@@ -651,6 +651,7 @@ type_size(const struct glsl_type *type)
         return 1;
       }
    case GLSL_TYPE_ARRAY:
+      assert(type->length > 0);
       return type_size(type->fields.array) * type->length;
    case GLSL_TYPE_STRUCT:
       size = 0;

I added two piglit test cases that reproduce this and a related bug.  The following patch series should fix both.

I'm leaving the bug open but resetting the version to 7.10 until the patches are cherry picked.

commit 0f4b2a0a23650d8f773c53d84cb2ead1f6d4fc8e
Author: Ian Romanick <ian.d.romanick@intel.com>
Date:   Tue Jan 25 12:06:18 2011 -0800

    linker: Propagate max_array_access while linking functions
    
    Update the max_array_access of a global as functions that use that
    global are pulled into the linked shader.
    
    Fixes piglit test glsl-fs-implicit-array-size-01 and bugzilla #33219.
    
    NOTE: This is a candidate for the 7.9 and 7.10 branches.

commit c87e9ef4d291b3fc18f7af2c7a7646b9a860f4af
Author: Ian Romanick <ian.d.romanick@intel.com>
Date:   Tue Jan 25 12:04:08 2011 -0800

    linker: Set sizes for non-global arrays as well
    
    Previously only global arrays with implicit sizes would be patched.
    This causes all arrays that are actually accessed to be sized.
    
    Fixes piglit test glsl-fs-implicit-array-size-02.
    
    NOTE: This is a candidate for the 7.9 and 7.10 branches.

Fixed by 2b115017 (7.9) and 3370f9b6 (7.10).

verified