Not a typo. Replace the default builder with one of bogus width to
catch cases in which optimization passes assume that the default
dispatch width is good enough. The execution controls of instructions
emitted during optimization should in general match the original code
that is being manipulated. Many of the problems fixed in this series
were caught by the assertions introduced in this patch.
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
* Ideally optimization passes wouldn't be part of the visitor so they
* wouldn't have access to bld at all, but they do, so just in case some
* pass forgets to ask for a location explicitly set it to NULL here to
- * make it trip.
+ * make it trip. The dispatch width is initialized to a bogus value to
+ * make sure that optimizations set the execution controls explicitly to
+ * match the code they are manipulating instead of relying on the defaults.
*/
- bld = bld.at(NULL, NULL);
+ bld = fs_builder(this, 64);
split_virtual_grfs();
dst_reg
vgrf(enum brw_reg_type type, unsigned n = 1) const
{
+ assert(dispatch_width() <= 32);
+
if (n > 0)
return dst_reg(GRF, shader->alloc.allocate(
DIV_ROUND_UP(n * type_sz(type) * dispatch_width(),
instruction *
emit(instruction *inst) const
{
+ assert(inst->exec_size <= 32);
assert(inst->exec_size == dispatch_width() ||
force_writemask_all);
assert(_group == 0 || _group == 8);