On 04/04/2012 07:33 AM, Vadim Girlin wrote:
Use the hash of the variable name instead of the pointer value.
Signed-off-by: Vadim Girlin<vadimgir...@gmail.com>
---
v2: Added the comment for the hash function as suggested by Vincent Lejeune.
Added const qualifier and modified the cast as suggested by Tolga Dalman.
src/glsl/lower_output_reads.cpp | 15 ++++++++++++++-
1 files changed, 14 insertions(+), 1 deletions(-)
diff --git a/src/glsl/lower_output_reads.cpp b/src/glsl/lower_output_reads.cpp
index 415b541..0427f0d 100644
--- a/src/glsl/lower_output_reads.cpp
+++ b/src/glsl/lower_output_reads.cpp
@@ -54,11 +54,24 @@ public:
virtual ir_visitor_status visit_leave(class ir_function_signature *);
};
+/**
+ * Hash function for the output variables - computes the hash of the name.
+ * NOTE: We're using the name string to ensure that the hash doesn't depend
+ * on any random factors, otherwise the output_read_remover could produce
+ * the random order of the assignments.
+ */
+static unsigned
+hash_table_var_hash(const void *key)
+{
+ const ir_variable * var = static_cast<const ir_variable *>(key);
+ return hash_table_string_hash(var->name);
We don't usually use the name as the hashkey because you can have
multiple distinct variables with the same name. These can be variables
from different scopes or compiler-generated names (though we try to
generate unique names in that case). Since this pass only applies to
shader outputs, I don't think this problem can occur. Shader outputs
should have unique names.
My main concern is that someone may see this code in the future and copy
it in a situation where it is not safe. With the addition of some
comment that describes this,
Reviewed-by: Ian Romanick <ian.d.roman...@intel.com>
+}
+
output_read_remover::output_read_remover()
{
mem_ctx = ralloc_context(NULL);
replacements =
- hash_table_ctor(0, hash_table_pointer_hash, hash_table_pointer_compare);
+ hash_table_ctor(0, hash_table_var_hash, hash_table_pointer_compare);
}
output_read_remover::~output_read_remover()
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
