i965/gen5: Fix grf_used calculation for 16-wide.

If we happened to allocate a texture result (or other vector) to the
highest hardware register slot, and we were in 16-wide, we would
under-count the registers used and potentially wrap around to g0 if
that allocation crossed a 16-register block boundary.  Bad rendering
and hangs ensued.

Tested-by: Ian Romanick <idr@freedesktop.org>

1 files changed, 4 insertions, 5 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp b/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp
index f88b1316775..b4689d2c293 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp
@@ -101,7 +101,6 @@ fs_visitor::assign_regs()
     * for reg_width == 2.
     */
    int reg_width = c->dispatch_width / 8;
-   int last_grf = 0;
    int hw_reg_mapping[this->virtual_grf_next + 1];
    int first_assigned_grf = ALIGN(this->first_non_payload_grf, reg_width);
    int base_reg_count = (BRW_MAX_GRF - first_assigned_grf) / reg_width;
@@ -263,6 +262,7 @@ fs_visitor::assign_regs()
     * regs in the register classes back down to real hardware reg
     * numbers.
     */
+   this->grf_used = first_assigned_grf;
    hw_reg_mapping[0] = 0; /* unused */
    for (int i = 1; i < this->virtual_grf_next; i++) {
       int reg = ra_get_node_reg(g, i);
@@ -278,8 +278,9 @@ fs_visitor::assign_regs()
 
       assert(hw_reg >= 0);
       hw_reg_mapping[i] = first_assigned_grf + hw_reg * reg_width;
-      last_grf = MAX2(last_grf,
-		      hw_reg_mapping[i] + this->virtual_grf_sizes[i] - 1);
+      this->grf_used = MAX2(this->grf_used,
+			    hw_reg_mapping[i] + this->virtual_grf_sizes[i] *
+			    reg_width);
    }
 
    foreach_iter(exec_list_iterator, iter, this->instructions) {
@@ -290,8 +291,6 @@ fs_visitor::assign_regs()
       assign_reg(hw_reg_mapping, &inst->src[1], reg_width);
    }
 
-   this->grf_used = last_grf + reg_width;
-
    ralloc_free(g);
    ralloc_free(regs);