[Mesa-dev] [PATCH 3/5] intel/fs: Cap dst-aligned region stride to maximum representable hstride value.

[Francisco Jerez]

This is required in combination with the following commit, because
otherwise if a source region with an extended 8+ stride is present in
the instruction (which we're about to declare legal) we'll end up
emitting code that attempts to write to such a region, even though
strides greater than four are still illegal for the destination.
---
 src/intel/compiler/brw_fs_lower_regioning.cpp | 20 ++++++++++++++-----
 1 file changed, 15 insertions(+), 5 deletions(-)

diff --git a/src/intel/compiler/brw_fs_lower_regioning.cpp b/src/intel/compiler/brw_fs_lower_regioning.cpp
index 6a3c23892b4..b86e95ed9eb 100644
--- a/src/intel/compiler/brw_fs_lower_regioning.cpp
+++ b/src/intel/compiler/brw_fs_lower_regioning.cpp
@@ -71,15 +71,25 @@ namespace {
           !is_byte_raw_mov(inst)) {
          return get_exec_type_size(inst);
       } else {
-         unsigned stride = inst->dst.stride * type_sz(inst->dst.type);
+         /* Calculate the maximum byte stride and the minimum type size across
+          * all source and destination operands.
+          */
+         unsigned max_stride = inst->dst.stride * type_sz(inst->dst.type);
+         unsigned min_size = type_sz(inst->dst.type);
 
          for (unsigned i = 0; i < inst->sources; i++) {
-            if (!is_uniform(inst->src[i]) && !inst->is_control_source(i))
-               stride = MAX2(stride, inst->src[i].stride *
-                             type_sz(inst->src[i].type));
+            if (!is_uniform(inst->src[i]) && !inst->is_control_source(i)) {
+               max_stride = MAX2(max_stride, inst->src[i].stride *
+                                 type_sz(inst->src[i].type));
+               min_size = MIN2(min_size, type_sz(inst->src[i].type));
+            }
          }
 
-         return stride;
+         /* Attempt to use the largest byte stride among all present operands,
+          * but never exceed a stride of 4 since that would lead to illegal
+          * destination regions during lowering.
+          */
+         return MIN2(max_stride, 4 * min_size);
       }
    }
 
-- 
2.19.2